Import OpenSSL 1.1.1h.vendor/openssl/1.1.1h

13 files changed, 212 insertions, 2378 deletions

diff --git a/crypto/ec/asm/ecp_nistz256-armv4.pl b/crypto/ec/asm/ecp_nistz256-armv4.pl
index ea538c0698d5..fa833ce6aaf3 100755
--- a/crypto/ec/asm/ecp_nistz256-armv4.pl
+++ b/crypto/ec/asm/ecp_nistz256-armv4.pl
@@ -1517,9 +1517,9 @@ ecp_nistz256_point_add:
 	ldr	$t2,[sp,#32*18+12]	@ ~is_equal(S1,S2)
 	mvn	$t0,$t0			@ -1/0 -> 0/-1
 	mvn	$t1,$t1			@ -1/0 -> 0/-1
-	orr	$a0,$t0
-	orr	$a0,$t1
-	orrs	$a0,$t2			@ set flags
+	orr	$a0,$a0,$t0
+	orr	$a0,$a0,$t1
+	orrs	$a0,$a0,$t2		@ set flags
 
 	@ if(~is_equal(U1,U2) | in1infty | in2infty | ~is_equal(S1,S2))
 	bne	.Ladd_proceed
diff --git a/crypto/ec/asm/ecp_nistz256-avx2.pl b/crypto/ec/asm/ecp_nistz256-avx2.pl
deleted file mode 100755
index 5071d09ac2ec..000000000000
--- a/crypto/ec/asm/ecp_nistz256-avx2.pl
+++ /dev/null
@@ -1,2080 +0,0 @@
-#! /usr/bin/env perl
-# Copyright 2014-2020 The OpenSSL Project Authors. All Rights Reserved.
-# Copyright (c) 2014, Intel Corporation. All Rights Reserved.
-#
-# Licensed under the OpenSSL license (the "License").  You may not use
-# this file except in compliance with the License.  You can obtain a copy
-# in the file LICENSE in the source distribution or at
-# https://www.openssl.org/source/license.html
-#
-# Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1)
-# (1) Intel Corporation, Israel Development Center, Haifa, Israel
-# (2) University of Haifa, Israel
-#
-# Reference:
-# S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with
-#                          256 Bit Primes"
-
-$flavour = shift;
-$output  = shift;
-if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-
-$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
-$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
-( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
-( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
-die "can't locate x86_64-xlate.pl";
-
-open OUT,"| \"$^X\" $xlate $flavour $output";
-*STDOUT=*OUT;
-
-if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
-		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
-	$avx = ($1>=2.19) + ($1>=2.22);
-	$addx = ($1>=2.23);
-}
-
-if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
-	    `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
-	$avx = ($1>=2.09) + ($1>=2.10);
-	$addx = ($1>=2.10);
-}
-
-if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
-	    `ml64 2>&1` =~ /Version ([0-9]+)\./) {
-	$avx = ($1>=10) + ($1>=11);
-	$addx = ($1>=12);
-}
-
-if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|based on LLVM) ([0-9]+)\.([0-9]+)/) {
-	my $ver = $2 + $3/100.0;	# 3.1->3.01, 3.10->3.10
-	$avx = ($ver>=3.0) + ($ver>=3.01);
-	$addx = ($ver>=3.03);
-}
-
-if ($avx>=2) {{
-$digit_size = "\$29";
-$n_digits = "\$9";
-
-$code.=<<___;
-.text
-
-.align 64
-.LAVX2_AND_MASK:
-.LAVX2_POLY:
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x00040000, 0x00040000, 0x00040000, 0x00040000
-.quad 0x1fe00000, 0x1fe00000, 0x1fe00000, 0x1fe00000
-.quad 0x00ffffff, 0x00ffffff, 0x00ffffff, 0x00ffffff
-
-.LAVX2_POLY_x2:
-.quad 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC
-.quad 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC
-.quad 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC
-.quad 0x400007FC, 0x400007FC, 0x400007FC, 0x400007FC
-.quad 0x3FFFFFFE, 0x3FFFFFFE, 0x3FFFFFFE, 0x3FFFFFFE
-.quad 0x3FFFFFFE, 0x3FFFFFFE, 0x3FFFFFFE, 0x3FFFFFFE
-.quad 0x400FFFFE, 0x400FFFFE, 0x400FFFFE, 0x400FFFFE
-.quad 0x7F7FFFFE, 0x7F7FFFFE, 0x7F7FFFFE, 0x7F7FFFFE
-.quad 0x03FFFFFC, 0x03FFFFFC, 0x03FFFFFC, 0x03FFFFFC
-
-.LAVX2_POLY_x8:
-.quad 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8
-.quad 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8
-.quad 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8, 0xFFFFFFF8
-.quad 0x80000FF8, 0x80000FF8, 0x80000FF8, 0x80000FF8
-.quad 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC
-.quad 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC, 0x7FFFFFFC
-.quad 0x801FFFFC, 0x801FFFFC, 0x801FFFFC, 0x801FFFFC
-.quad 0xFEFFFFFC, 0xFEFFFFFC, 0xFEFFFFFC, 0xFEFFFFFC
-.quad 0x07FFFFF8, 0x07FFFFF8, 0x07FFFFF8, 0x07FFFFF8
-
-.LONE:
-.quad 0x00000020, 0x00000020, 0x00000020, 0x00000020
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x1fffc000, 0x1fffc000, 0x1fffc000, 0x1fffc000
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1f7fffff, 0x1f7fffff, 0x1f7fffff, 0x1f7fffff
-.quad 0x03ffffff, 0x03ffffff, 0x03ffffff, 0x03ffffff
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-
-# RR = 2^266 mod p in AVX2 format, to transform from the native OpenSSL
-# Montgomery form (*2^256) to our format (*2^261)
-
-.LTO_MONT_AVX2:
-.quad 0x00000400, 0x00000400, 0x00000400, 0x00000400
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x1ff80000, 0x1ff80000, 0x1ff80000, 0x1ff80000
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x0fffffff, 0x0fffffff, 0x0fffffff, 0x0fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x00000003, 0x00000003, 0x00000003, 0x00000003
-
-.LFROM_MONT_AVX2:
-.quad 0x00000001, 0x00000001, 0x00000001, 0x00000001
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-.quad 0x1ffffe00, 0x1ffffe00, 0x1ffffe00, 0x1ffffe00
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1fffffff, 0x1fffffff, 0x1fffffff, 0x1fffffff
-.quad 0x1ffbffff, 0x1ffbffff, 0x1ffbffff, 0x1ffbffff
-.quad 0x001fffff, 0x001fffff, 0x001fffff, 0x001fffff
-.quad 0x00000000, 0x00000000, 0x00000000, 0x00000000
-
-.LIntOne:
-.long 1,1,1,1,1,1,1,1
-___
-
-{
-# This function receives a pointer to an array of four affine points
-# (X, Y, <1>) and rearranges the data for AVX2 execution, while
-# converting it to 2^29 radix redundant form
-
-my ($X0,$X1,$X2,$X3, $Y0,$Y1,$Y2,$Y3,
-    $T0,$T1,$T2,$T3, $T4,$T5,$T6,$T7)=map("%ymm$_",(0..15));
-
-$code.=<<___;
-.globl	ecp_nistz256_avx2_transpose_convert
-.type	ecp_nistz256_avx2_transpose_convert,\@function,2
-.align 64
-ecp_nistz256_avx2_transpose_convert:
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-8-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	# Load the data
-	vmovdqa		32*0(%rsi), $X0
-	lea		112(%rsi), %rax		# size optimization
-	vmovdqa		32*1(%rsi), $Y0
-	lea		.LAVX2_AND_MASK(%rip), %rdx
-	vmovdqa		32*2(%rsi), $X1
-	vmovdqa		32*3(%rsi), $Y1
-	vmovdqa		32*4-112(%rax), $X2
-	vmovdqa		32*5-112(%rax), $Y2
-	vmovdqa		32*6-112(%rax), $X3
-	vmovdqa		32*7-112(%rax), $Y3
-
-	# Transpose X and Y independently
-	vpunpcklqdq	$X1, $X0, $T0		# T0 = [B2 A2 B0 A0]
-	vpunpcklqdq	$X3, $X2, $T1		# T1 = [D2 C2 D0 C0]
-	vpunpckhqdq	$X1, $X0, $T2		# T2 = [B3 A3 B1 A1]
-	vpunpckhqdq	$X3, $X2, $T3		# T3 = [D3 C3 D1 C1]
-
-	vpunpcklqdq	$Y1, $Y0, $T4
-	vpunpcklqdq	$Y3, $Y2, $T5
-	vpunpckhqdq	$Y1, $Y0, $T6
-	vpunpckhqdq	$Y3, $Y2, $T7
-
-	vperm2i128	\$0x20, $T1, $T0, $X0	# X0 = [D0 C0 B0 A0]
-	vperm2i128	\$0x20, $T3, $T2, $X1	# X1 = [D1 C1 B1 A1]
-	vperm2i128	\$0x31, $T1, $T0, $X2	# X2 = [D2 C2 B2 A2]
-	vperm2i128	\$0x31, $T3, $T2, $X3	# X3 = [D3 C3 B3 A3]
-
-	vperm2i128	\$0x20, $T5, $T4, $Y0
-	vperm2i128	\$0x20, $T7, $T6, $Y1
-	vperm2i128	\$0x31, $T5, $T4, $Y2
-	vperm2i128	\$0x31, $T7, $T6, $Y3
-	vmovdqa		(%rdx), $T7
-
-	vpand		(%rdx), $X0, $T0	# out[0] = in[0] & mask;
-	vpsrlq		\$29, $X0, $X0
-	vpand		$T7, $X0, $T1		# out[1] = (in[0] >> shift) & mask;
-	vpsrlq		\$29, $X0, $X0
-	vpsllq		\$6, $X1, $T2
-	vpxor		$X0, $T2, $T2
-	vpand		$T7, $T2, $T2		# out[2] = ((in[0] >> (shift*2)) ^ (in[1] << (64-shift*2))) & mask;
-	vpsrlq		\$23, $X1, $X1
-	vpand		$T7, $X1, $T3		# out[3] = (in[1] >> ((shift*3)%64)) & mask;
-	vpsrlq		\$29, $X1, $X1
-	vpsllq		\$12, $X2, $T4
-	vpxor		$X1, $T4, $T4
-	vpand		$T7, $T4, $T4		# out[4] = ((in[1] >> ((shift*4)%64)) ^ (in[2] << (64*2-shift*4))) & mask;
-	vpsrlq		\$17, $X2, $X2
-	vpand		$T7, $X2, $T5		# out[5] = (in[2] >> ((shift*5)%64)) & mask;
-	vpsrlq		\$29, $X2, $X2
-	vpsllq		\$18, $X3, $T6
-	vpxor		$X2, $T6, $T6
-	vpand		$T7, $T6, $T6		# out[6] = ((in[2] >> ((shift*6)%64)) ^ (in[3] << (64*3-shift*6))) & mask;
-	vpsrlq		\$11, $X3, $X3
-	 vmovdqa	$T0, 32*0(%rdi)
-	 lea		112(%rdi), %rax		# size optimization
-	vpand		$T7, $X3, $T0		# out[7] = (in[3] >> ((shift*7)%64)) & mask;
-	vpsrlq		\$29, $X3, $X3		# out[8] = (in[3] >> ((shift*8)%64)) & mask;
-
-	vmovdqa		$T1, 32*1(%rdi)
-	vmovdqa		$T2, 32*2(%rdi)
-	vmovdqa		$T3, 32*3(%rdi)
-	vmovdqa		$T4, 32*4-112(%rax)
-	vmovdqa		$T5, 32*5-112(%rax)
-	vmovdqa		$T6, 32*6-112(%rax)
-	vmovdqa		$T0, 32*7-112(%rax)
-	vmovdqa		$X3, 32*8-112(%rax)
-	lea		448(%rdi), %rax		# size optimization
-
-	vpand		$T7, $Y0, $T0		# out[0] = in[0] & mask;
-	vpsrlq		\$29, $Y0, $Y0
-	vpand		$T7, $Y0, $T1		# out[1] = (in[0] >> shift) & mask;
-	vpsrlq		\$29, $Y0, $Y0
-	vpsllq		\$6, $Y1, $T2
-	vpxor		$Y0, $T2, $T2
-	vpand		$T7, $T2, $T2		# out[2] = ((in[0] >> (shift*2)) ^ (in[1] << (64-shift*2))) & mask;
-	vpsrlq		\$23, $Y1, $Y1
-	vpand		$T7, $Y1, $T3		# out[3] = (in[1] >> ((shift*3)%64)) & mask;
-	vpsrlq		\$29, $Y1, $Y1
-	vpsllq		\$12, $Y2, $T4
-	vpxor		$Y1, $T4, $T4
-	vpand		$T7, $T4, $T4		# out[4] = ((in[1] >> ((shift*4)%64)) ^ (in[2] << (64*2-shift*4))) & mask;
-	vpsrlq		\$17, $Y2, $Y2
-	vpand		$T7, $Y2, $T5		# out[5] = (in[2] >> ((shift*5)%64)) & mask;
-	vpsrlq		\$29, $Y2, $Y2
-	vpsllq		\$18, $Y3, $T6
-	vpxor		$Y2, $T6, $T6
-	vpand		$T7, $T6, $T6		# out[6] = ((in[2] >> ((shift*6)%64)) ^ (in[3] << (64*3-shift*6))) & mask;
-	vpsrlq		\$11, $Y3, $Y3
-	 vmovdqa	$T0, 32*9-448(%rax)
-	vpand		$T7, $Y3, $T0		# out[7] = (in[3] >> ((shift*7)%64)) & mask;
-	vpsrlq		\$29, $Y3, $Y3		# out[8] = (in[3] >> ((shift*8)%64)) & mask;
-
-	vmovdqa		$T1, 32*10-448(%rax)
-	vmovdqa		$T2, 32*11-448(%rax)
-	vmovdqa		$T3, 32*12-448(%rax)
-	vmovdqa		$T4, 32*13-448(%rax)
-	vmovdqa		$T5, 32*14-448(%rax)
-	vmovdqa		$T6, 32*15-448(%rax)
-	vmovdqa		$T0, 32*16-448(%rax)
-	vmovdqa		$Y3, 32*17-448(%rax)
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	16*0(%rsp), %xmm6
-	movaps	16*1(%rsp), %xmm7
-	movaps	16*2(%rsp), %xmm8
-	movaps	16*3(%rsp), %xmm9
-	movaps	16*4(%rsp), %xmm10
-	movaps	16*5(%rsp), %xmm11
-	movaps	16*6(%rsp), %xmm12
-	movaps	16*7(%rsp), %xmm13
-	movaps	16*8(%rsp), %xmm14
-	movaps	16*9(%rsp), %xmm15
-	lea	8+16*10(%rsp), %rsp
-___
-$code.=<<___;
-	ret
-.size	ecp_nistz256_avx2_transpose_convert,.-ecp_nistz256_avx2_transpose_convert
-___
-}
-{
-################################################################################
-# This function receives a pointer to an array of four AVX2 formatted points
-# (X, Y, Z) convert the data to normal representation, and rearranges the data
-
-my ($D0,$D1,$D2,$D3, $D4,$D5,$D6,$D7, $D8)=map("%ymm$_",(0..8));
-my ($T0,$T1,$T2,$T3, $T4,$T5,$T6)=map("%ymm$_",(9..15));
-
-$code.=<<___;
-
-.globl	ecp_nistz256_avx2_convert_transpose_back
-.type	ecp_nistz256_avx2_convert_transpose_back,\@function,2
-.align	32
-ecp_nistz256_avx2_convert_transpose_back:
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-8-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	mov	\$3, %ecx
-
-.Lconv_loop:
-	vmovdqa		32*0(%rsi), $D0
-	lea		160(%rsi), %rax		# size optimization
-	vmovdqa		32*1(%rsi), $D1
-	vmovdqa		32*2(%rsi), $D2
-	vmovdqa		32*3(%rsi), $D3
-	vmovdqa		32*4-160(%rax), $D4
-	vmovdqa		32*5-160(%rax), $D5
-	vmovdqa		32*6-160(%rax), $D6
-	vmovdqa		32*7-160(%rax), $D7
-	vmovdqa		32*8-160(%rax), $D8
-
-	vpsllq		\$29, $D1, $D1
-	vpsllq		\$58, $D2, $T0
-	vpaddq		$D1, $D0, $D0
-	vpaddq		$T0, $D0, $D0		# out[0] = (in[0]) ^ (in[1] << shift*1) ^ (in[2] << shift*2);
-
-	vpsrlq		\$6, $D2, $D2
-	vpsllq		\$23, $D3, $D3
-	vpsllq		\$52, $D4, $T1
-	vpaddq		$D2, $D3, $D3
-	vpaddq		$D3, $T1, $D1		# out[1] = (in[2] >> (64*1-shift*2)) ^ (in[3] << shift*3%64) ^ (in[4] << shift*4%64);
-
-	vpsrlq		\$12, $D4, $D4
-	vpsllq		\$17, $D5, $D5
-	vpsllq		\$46, $D6, $T2
-	vpaddq		$D4, $D5, $D5
-	vpaddq		$D5, $T2, $D2		# out[2] = (in[4] >> (64*2-shift*4)) ^ (in[5] << shift*5%64) ^ (in[6] << shift*6%64);
-
-	vpsrlq		\$18, $D6, $D6
-	vpsllq		\$11, $D7, $D7
-	vpsllq		\$40, $D8, $T3
-	vpaddq		$D6, $D7, $D7
-	vpaddq		$D7, $T3, $D3		# out[3] = (in[6] >> (64*3-shift*6)) ^ (in[7] << shift*7%64) ^ (in[8] << shift*8%64);
-
-	vpunpcklqdq	$D1, $D0, $T0		# T0 = [B2 A2 B0 A0]
-	vpunpcklqdq	$D3, $D2, $T1		# T1 = [D2 C2 D0 C0]
-	vpunpckhqdq	$D1, $D0, $T2		# T2 = [B3 A3 B1 A1]
-	vpunpckhqdq	$D3, $D2, $T3		# T3 = [D3 C3 D1 C1]
-
-	vperm2i128	\$0x20, $T1, $T0, $D0	# X0 = [D0 C0 B0 A0]
-	vperm2i128	\$0x20, $T3, $T2, $D1	# X1 = [D1 C1 B1 A1]
-	vperm2i128	\$0x31, $T1, $T0, $D2	# X2 = [D2 C2 B2 A2]
-	vperm2i128	\$0x31, $T3, $T2, $D3	# X3 = [D3 C3 B3 A3]
-
-	vmovdqa		$D0, 32*0(%rdi)
-	vmovdqa		$D1, 32*3(%rdi)
-	vmovdqa		$D2, 32*6(%rdi)
-	vmovdqa		$D3, 32*9(%rdi)
-
-	lea		32*9(%rsi), %rsi
-	lea		32*1(%rdi), %rdi
-
-	dec	%ecx
-	jnz	.Lconv_loop
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	16*0(%rsp), %xmm6
-	movaps	16*1(%rsp), %xmm7
-	movaps	16*2(%rsp), %xmm8
-	movaps	16*3(%rsp), %xmm9
-	movaps	16*4(%rsp), %xmm10
-	movaps	16*5(%rsp), %xmm11
-	movaps	16*6(%rsp), %xmm12
-	movaps	16*7(%rsp), %xmm13
-	movaps	16*8(%rsp), %xmm14
-	movaps	16*9(%rsp), %xmm15
-	lea	8+16*10(%rsp), %rsp
-___
-$code.=<<___;
-	ret
-.size	ecp_nistz256_avx2_convert_transpose_back,.-ecp_nistz256_avx2_convert_transpose_back
-___
-}
-{
-my ($r_ptr,$a_ptr,$b_ptr,$itr)=("%rdi","%rsi","%rdx","%ecx");
-my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4,$ACC5,$ACC6,$ACC7,$ACC8)=map("%ymm$_",(0..8));
-my ($B,$Y,$T0,$AND_MASK,$OVERFLOW)=map("%ymm$_",(9..13));
-
-sub NORMALIZE {
-my $ret=<<___;
-	vpsrlq		$digit_size, $ACC0, $T0
-	vpand		$AND_MASK, $ACC0, $ACC0
-	vpaddq		$T0, $ACC1, $ACC1
-
-	vpsrlq		$digit_size, $ACC1, $T0
-	vpand		$AND_MASK, $ACC1, $ACC1
-	vpaddq		$T0, $ACC2, $ACC2
-
-	vpsrlq		$digit_size, $ACC2, $T0
-	vpand		$AND_MASK, $ACC2, $ACC2
-	vpaddq		$T0, $ACC3, $ACC3
-
-	vpsrlq		$digit_size, $ACC3, $T0
-	vpand		$AND_MASK, $ACC3, $ACC3
-	vpaddq		$T0, $ACC4, $ACC4
-
-	vpsrlq		$digit_size, $ACC4, $T0
-	vpand		$AND_MASK, $ACC4, $ACC4
-	vpaddq		$T0, $ACC5, $ACC5
-
-	vpsrlq		$digit_size, $ACC5, $T0
-	vpand		$AND_MASK, $ACC5, $ACC5
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpsrlq		$digit_size, $ACC6, $T0
-	vpand		$AND_MASK, $ACC6, $ACC6
-	vpaddq		$T0, $ACC7, $ACC7
-
-	vpsrlq		$digit_size, $ACC7, $T0
-	vpand		$AND_MASK, $ACC7, $ACC7
-	vpaddq		$T0, $ACC8, $ACC8
-	#vpand		$AND_MASK, $ACC8, $ACC8
-___
-    $ret;
-}
-
-sub STORE {
-my $ret=<<___;
-	vmovdqa		$ACC0, 32*0(%rdi)
-	lea		160(%rdi), %rax		# size optimization
-	vmovdqa		$ACC1, 32*1(%rdi)
-	vmovdqa		$ACC2, 32*2(%rdi)
-	vmovdqa		$ACC3, 32*3(%rdi)
-	vmovdqa		$ACC4, 32*4-160(%rax)
-	vmovdqa		$ACC5, 32*5-160(%rax)
-	vmovdqa		$ACC6, 32*6-160(%rax)
-	vmovdqa		$ACC7, 32*7-160(%rax)
-	vmovdqa		$ACC8, 32*8-160(%rax)
-___
-    $ret;
-}
-
-$code.=<<___;
-.type	avx2_normalize,\@abi-omnipotent
-.align	32
-avx2_normalize:
-	vpsrlq		$digit_size, $ACC0, $T0
-	vpand		$AND_MASK, $ACC0, $ACC0
-	vpaddq		$T0, $ACC1, $ACC1
-
-	vpsrlq		$digit_size, $ACC1, $T0
-	vpand		$AND_MASK, $ACC1, $ACC1
-	vpaddq		$T0, $ACC2, $ACC2
-
-	vpsrlq		$digit_size, $ACC2, $T0
-	vpand		$AND_MASK, $ACC2, $ACC2
-	vpaddq		$T0, $ACC3, $ACC3
-
-	vpsrlq		$digit_size, $ACC3, $T0
-	vpand		$AND_MASK, $ACC3, $ACC3
-	vpaddq		$T0, $ACC4, $ACC4
-
-	vpsrlq		$digit_size, $ACC4, $T0
-	vpand		$AND_MASK, $ACC4, $ACC4
-	vpaddq		$T0, $ACC5, $ACC5
-
-	vpsrlq		$digit_size, $ACC5, $T0
-	vpand		$AND_MASK, $ACC5, $ACC5
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpsrlq		$digit_size, $ACC6, $T0
-	vpand		$AND_MASK, $ACC6, $ACC6
-	vpaddq		$T0, $ACC7, $ACC7
-
-	vpsrlq		$digit_size, $ACC7, $T0
-	vpand		$AND_MASK, $ACC7, $ACC7
-	vpaddq		$T0, $ACC8, $ACC8
-	#vpand		$AND_MASK, $ACC8, $ACC8
-
-	ret
-.size	avx2_normalize,.-avx2_normalize
-
-.type	avx2_normalize_n_store,\@abi-omnipotent
-.align	32
-avx2_normalize_n_store:
-	vpsrlq		$digit_size, $ACC0, $T0
-	vpand		$AND_MASK, $ACC0, $ACC0
-	vpaddq		$T0, $ACC1, $ACC1
-
-	vpsrlq		$digit_size, $ACC1, $T0
-	vpand		$AND_MASK, $ACC1, $ACC1
-	 vmovdqa	$ACC0, 32*0(%rdi)
-	 lea		160(%rdi), %rax		# size optimization
-	vpaddq		$T0, $ACC2, $ACC2
-
-	vpsrlq		$digit_size, $ACC2, $T0
-	vpand		$AND_MASK, $ACC2, $ACC2
-	 vmovdqa	$ACC1, 32*1(%rdi)
-	vpaddq		$T0, $ACC3, $ACC3
-
-	vpsrlq		$digit_size, $ACC3, $T0
-	vpand		$AND_MASK, $ACC3, $ACC3
-	 vmovdqa	$ACC2, 32*2(%rdi)
-	vpaddq		$T0, $ACC4, $ACC4
-
-	vpsrlq		$digit_size, $ACC4, $T0
-	vpand		$AND_MASK, $ACC4, $ACC4
-	 vmovdqa	$ACC3, 32*3(%rdi)
-	vpaddq		$T0, $ACC5, $ACC5
-
-	vpsrlq		$digit_size, $ACC5, $T0
-	vpand		$AND_MASK, $ACC5, $ACC5
-	 vmovdqa	$ACC4, 32*4-160(%rax)
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpsrlq		$digit_size, $ACC6, $T0
-	vpand		$AND_MASK, $ACC6, $ACC6
-	 vmovdqa	$ACC5, 32*5-160(%rax)
-	vpaddq		$T0, $ACC7, $ACC7
-
-	vpsrlq		$digit_size, $ACC7, $T0
-	vpand		$AND_MASK, $ACC7, $ACC7
-	 vmovdqa	$ACC6, 32*6-160(%rax)
-	vpaddq		$T0, $ACC8, $ACC8
-	#vpand		$AND_MASK, $ACC8, $ACC8
-	 vmovdqa	$ACC7, 32*7-160(%rax)
-	 vmovdqa	$ACC8, 32*8-160(%rax)
-
-	ret
-.size	avx2_normalize_n_store,.-avx2_normalize_n_store
-
-################################################################################
-# void avx2_mul_x4(void* RESULTx4, void *Ax4, void *Bx4);
-.type	avx2_mul_x4,\@abi-omnipotent
-.align	32
-avx2_mul_x4:
-	lea	.LAVX2_POLY(%rip), %rax
-
-	vpxor	$ACC0, $ACC0, $ACC0
-	vpxor	$ACC1, $ACC1, $ACC1
-	vpxor	$ACC2, $ACC2, $ACC2
-	vpxor	$ACC3, $ACC3, $ACC3
-	vpxor	$ACC4, $ACC4, $ACC4
-	vpxor	$ACC5, $ACC5, $ACC5
-	vpxor	$ACC6, $ACC6, $ACC6
-	vpxor	$ACC7, $ACC7, $ACC7
-
-	vmovdqa	32*7(%rax), %ymm14
-	vmovdqa	32*8(%rax), %ymm15
-
-	mov	$n_digits, $itr
-	lea	-512($a_ptr), $a_ptr	# strategic bias to control u-op density
-	jmp	.Lavx2_mul_x4_loop
-
-.align	32
-.Lavx2_mul_x4_loop:
-	vmovdqa		32*0($b_ptr), $B
-	lea		32*1($b_ptr), $b_ptr
-
-	vpmuludq	32*0+512($a_ptr), $B, $T0
-	vpmuludq	32*1+512($a_ptr), $B, $OVERFLOW	# borrow $OVERFLOW
-	vpaddq		$T0, $ACC0, $ACC0
-	vpmuludq	32*2+512($a_ptr), $B, $T0
-	vpaddq		$OVERFLOW, $ACC1, $ACC1
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpmuludq	32*3+512($a_ptr), $B, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC2
-	vpmuludq	32*4+512($a_ptr), $B, $T0
-	vpaddq		$OVERFLOW, $ACC3, $ACC3
-	vpmuludq	32*5+512($a_ptr), $B, $OVERFLOW
-	vpaddq		$T0, $ACC4, $ACC4
-	vpmuludq	32*6+512($a_ptr), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*7+512($a_ptr), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	# Skip some multiplications, optimizing for the constant poly
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*8+512($a_ptr), $B, $ACC8
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	.byte		0x67
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $OVERFLOW
-	.byte		0x67
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $T0
-	vpaddq		$OVERFLOW, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC7, $ACC6
-	vpaddq		$OVERFLOW, $ACC8, $ACC7
-
-	dec	$itr
-	jnz	.Lavx2_mul_x4_loop
-
-	vpxor	$ACC8, $ACC8, $ACC8
-
-	ret
-.size	avx2_mul_x4,.-avx2_mul_x4
-
-# Function optimized for the constant 1
-################################################################################
-# void avx2_mul_by1_x4(void* RESULTx4, void *Ax4);
-.type	avx2_mul_by1_x4,\@abi-omnipotent
-.align	32
-avx2_mul_by1_x4:
-	lea	.LAVX2_POLY(%rip), %rax
-
-	vpxor	$ACC0, $ACC0, $ACC0
-	vpxor	$ACC1, $ACC1, $ACC1
-	vpxor	$ACC2, $ACC2, $ACC2
-	vpxor	$ACC3, $ACC3, $ACC3
-	vpxor	$ACC4, $ACC4, $ACC4
-	vpxor	$ACC5, $ACC5, $ACC5
-	vpxor	$ACC6, $ACC6, $ACC6
-	vpxor	$ACC7, $ACC7, $ACC7
-	vpxor	$ACC8, $ACC8, $ACC8
-
-	vmovdqa	32*3+.LONE(%rip), %ymm14
-	vmovdqa	32*7+.LONE(%rip), %ymm15
-
-	mov	$n_digits, $itr
-	jmp	.Lavx2_mul_by1_x4_loop
-
-.align	32
-.Lavx2_mul_by1_x4_loop:
-	vmovdqa		32*0($a_ptr), $B
-	.byte		0x48,0x8d,0xb6,0x20,0,0,0	# lea	32*1($a_ptr), $a_ptr
-
-	vpsllq		\$5, $B, $OVERFLOW
-	vpmuludq	%ymm14, $B, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC3
-	.byte		0x67
-	vpmuludq	$AND_MASK, $B, $T0
-	vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$T0, $ACC4, $ACC4
-	vpaddq		$T0, $ACC5, $ACC5
-	vpaddq		$T0, $ACC6, $ACC6
-	vpsllq		\$23, $B, $T0
-
-	.byte		0x67,0x67
-	vpmuludq	%ymm15, $B, $OVERFLOW
-	vpsubq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	vpaddq		$OVERFLOW, $ACC7, $ACC7
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	.byte		0x67,0x67
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $OVERFLOW
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	32*7(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC6, $ACC5
-	vpaddq		$T0, $ACC7, $ACC6
-	vpmuludq	32*8(%rax), $Y, $ACC7
-
-	dec	$itr
-	jnz	.Lavx2_mul_by1_x4_loop
-
-	ret
-.size	avx2_mul_by1_x4,.-avx2_mul_by1_x4
-
-################################################################################
-# void avx2_sqr_x4(void* RESULTx4, void *Ax4, void *Bx4);
-.type	avx2_sqr_x4,\@abi-omnipotent
-.align	32
-avx2_sqr_x4:
-	lea		.LAVX2_POLY(%rip), %rax
-
-	vmovdqa		32*7(%rax), %ymm14
-	vmovdqa		32*8(%rax), %ymm15
-
-	vmovdqa		32*0($a_ptr), $B
-	vmovdqa		32*1($a_ptr), $ACC1
-	vmovdqa		32*2($a_ptr), $ACC2
-	vmovdqa		32*3($a_ptr), $ACC3
-	vmovdqa		32*4($a_ptr), $ACC4
-	vmovdqa		32*5($a_ptr), $ACC5
-	vmovdqa		32*6($a_ptr), $ACC6
-	vmovdqa		32*7($a_ptr), $ACC7
-	vpaddq		$ACC1, $ACC1, $ACC1	# 2*$ACC0..7
-	vmovdqa		32*8($a_ptr), $ACC8
-	vpaddq		$ACC2, $ACC2, $ACC2
-	vmovdqa		$ACC1, 32*0(%rcx)
-	vpaddq		$ACC3, $ACC3, $ACC3
-	vmovdqa		$ACC2, 32*1(%rcx)
-	vpaddq		$ACC4, $ACC4, $ACC4
-	vmovdqa		$ACC3, 32*2(%rcx)
-	vpaddq		$ACC5, $ACC5, $ACC5
-	vmovdqa		$ACC4, 32*3(%rcx)
-	vpaddq		$ACC6, $ACC6, $ACC6
-	vmovdqa		$ACC5, 32*4(%rcx)
-	vpaddq		$ACC7, $ACC7, $ACC7
-	vmovdqa		$ACC6, 32*5(%rcx)
-	vpaddq		$ACC8, $ACC8, $ACC8
-	vmovdqa		$ACC7, 32*6(%rcx)
-	vmovdqa		$ACC8, 32*7(%rcx)
-
-	#itr		1
-	vpmuludq	$B, $B, $ACC0
-	vpmuludq	$B, $ACC1, $ACC1
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpmuludq	$B, $ACC2, $ACC2
-	vpmuludq	$B, $ACC3, $ACC3
-	vpmuludq	$B, $ACC4, $ACC4
-	vpmuludq	$B, $ACC5, $ACC5
-	vpmuludq	$B, $ACC6, $ACC6
-	 vpmuludq	$AND_MASK, $Y, $T0
-	vpmuludq	$B, $ACC7, $ACC7
-	vpmuludq	$B, $ACC8, $ACC8
-	 vmovdqa	32*1($a_ptr), $B
-
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		2
-	vpmuludq	$B, $B, $OVERFLOW
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpmuludq	32*1(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC1, $ACC1
-	vpmuludq	32*2(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC2
-	vpmuludq	32*3(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC3, $ACC3
-	vpmuludq	32*4(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC4, $ACC4
-	vpmuludq	32*5(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*2($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		3
-	vpmuludq	$B, $B, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpmuludq	32*2(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC2
-	vpmuludq	32*3(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC3, $ACC3
-	vpmuludq	32*4(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC4, $ACC4
-	vpmuludq	32*5(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*3($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		4
-	vpmuludq	$B, $B, $OVERFLOW
-	vpmuludq	32*3(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC3, $ACC3
-	vpmuludq	32*4(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC4, $ACC4
-	vpmuludq	32*5(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*4($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		5
-	vpmuludq	$B, $B, $T0
-	vpmuludq	32*4(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC4, $ACC4
-	vpmuludq	32*5(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*5($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3+.LAVX2_POLY(%rip), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		6
-	vpmuludq	$B, $B, $OVERFLOW
-	vpmuludq	32*5(%rcx), $B, $T0
-	vpaddq		$OVERFLOW, $ACC5, $ACC5
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*6($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		7
-	vpmuludq	$B, $B, $T0
-	vpmuludq	32*6(%rcx), $B, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC6
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*7($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		8
-	vpmuludq	$B, $B, $OVERFLOW
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	 vpaddq		$OVERFLOW, $ACC7, $ACC7
-	 vpmuludq	32*7(%rcx), $B, $ACC8
-	 vmovdqa	32*8($a_ptr), $B
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	 vpand		$AND_MASK, $ACC0, $Y
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	#itr		9
-	vpmuludq	$B, $B, $ACC8
-
-	vpmuludq	$AND_MASK, $Y, $T0
-	vpaddq		$T0, $ACC0, $OVERFLOW
-	vpsrlq		$digit_size, $OVERFLOW, $OVERFLOW
-	vpaddq		$T0, $ACC1, $ACC0
-	vpaddq		$T0, $ACC2, $ACC1
-	vpmuludq	32*3(%rax), $Y, $T0
-	vpaddq		$OVERFLOW, $ACC0, $ACC0
-	vpaddq		$T0, $ACC3, $ACC2
-	vmovdqa		$ACC4, $ACC3
-	vpsllq		\$18, $Y, $T0
-	vmovdqa		$ACC5, $ACC4
-	vpmuludq	%ymm14, $Y, $OVERFLOW
-	vpaddq		$T0, $ACC6, $ACC5
-	vpmuludq	%ymm15, $Y, $T0
-	vpaddq		$OVERFLOW, $ACC7, $ACC6
-	vpaddq		$T0, $ACC8, $ACC7
-
-	vpxor		$ACC8, $ACC8, $ACC8
-
-	ret
-.size	avx2_sqr_x4,.-avx2_sqr_x4
-
-################################################################################
-# void avx2_sub_x4(void* RESULTx4, void *Ax4, void *Bx4);
-.type	avx2_sub_x4,\@abi-omnipotent
-.align	32
-avx2_sub_x4:
-	vmovdqa	32*0($a_ptr), $ACC0
-	lea	160($a_ptr), $a_ptr
-	lea	.LAVX2_POLY_x8+128(%rip), %rax
-	lea	128($b_ptr), $b_ptr
-	vmovdqa	32*1-160($a_ptr), $ACC1
-	vmovdqa	32*2-160($a_ptr), $ACC2
-	vmovdqa	32*3-160($a_ptr), $ACC3
-	vmovdqa	32*4-160($a_ptr), $ACC4
-	vmovdqa	32*5-160($a_ptr), $ACC5
-	vmovdqa	32*6-160($a_ptr), $ACC6
-	vmovdqa	32*7-160($a_ptr), $ACC7
-	vmovdqa	32*8-160($a_ptr), $ACC8
-
-	vpaddq	32*0-128(%rax), $ACC0, $ACC0
-	vpaddq	32*1-128(%rax), $ACC1, $ACC1
-	vpaddq	32*2-128(%rax), $ACC2, $ACC2
-	vpaddq	32*3-128(%rax), $ACC3, $ACC3
-	vpaddq	32*4-128(%rax), $ACC4, $ACC4
-	vpaddq	32*5-128(%rax), $ACC5, $ACC5
-	vpaddq	32*6-128(%rax), $ACC6, $ACC6
-	vpaddq	32*7-128(%rax), $ACC7, $ACC7
-	vpaddq	32*8-128(%rax), $ACC8, $ACC8
-
-	vpsubq	32*0-128($b_ptr), $ACC0, $ACC0
-	vpsubq	32*1-128($b_ptr), $ACC1, $ACC1
-	vpsubq	32*2-128($b_ptr), $ACC2, $ACC2
-	vpsubq	32*3-128($b_ptr), $ACC3, $ACC3
-	vpsubq	32*4-128($b_ptr), $ACC4, $ACC4
-	vpsubq	32*5-128($b_ptr), $ACC5, $ACC5
-	vpsubq	32*6-128($b_ptr), $ACC6, $ACC6
-	vpsubq	32*7-128($b_ptr), $ACC7, $ACC7
-	vpsubq	32*8-128($b_ptr), $ACC8, $ACC8
-
-	ret
-.size	avx2_sub_x4,.-avx2_sub_x4
-
-.type	avx2_select_n_store,\@abi-omnipotent
-.align	32
-avx2_select_n_store:
-	vmovdqa	`8+32*9*8`(%rsp), $Y
-	vpor	`8+32*9*8+32`(%rsp), $Y, $Y
-
-	vpandn	$ACC0, $Y, $ACC0
-	vpandn	$ACC1, $Y, $ACC1
-	vpandn	$ACC2, $Y, $ACC2
-	vpandn	$ACC3, $Y, $ACC3
-	vpandn	$ACC4, $Y, $ACC4
-	vpandn	$ACC5, $Y, $ACC5
-	vpandn	$ACC6, $Y, $ACC6
-	vmovdqa	`8+32*9*8+32`(%rsp), $B
-	vpandn	$ACC7, $Y, $ACC7
-	vpandn	`8+32*9*8`(%rsp), $B, $B
-	vpandn	$ACC8, $Y, $ACC8
-
-	vpand	32*0(%rsi), $B, $T0
-	lea	160(%rsi), %rax
-	vpand	32*1(%rsi), $B, $Y
-	vpxor	$T0, $ACC0, $ACC0
-	vpand	32*2(%rsi), $B, $T0
-	vpxor	$Y, $ACC1, $ACC1
-	vpand	32*3(%rsi), $B, $Y
-	vpxor	$T0, $ACC2, $ACC2
-	vpand	32*4-160(%rax), $B, $T0
-	vpxor	$Y, $ACC3, $ACC3
-	vpand	32*5-160(%rax), $B, $Y
-	vpxor	$T0, $ACC4, $ACC4
-	vpand	32*6-160(%rax), $B, $T0
-	vpxor	$Y, $ACC5, $ACC5
-	vpand	32*7-160(%rax), $B, $Y
-	vpxor	$T0, $ACC6, $ACC6
-	vpand	32*8-160(%rax), $B, $T0
-	vmovdqa	`8+32*9*8+32`(%rsp), $B
-	vpxor	$Y, $ACC7, $ACC7
-
-	vpand	32*0(%rdx), $B, $Y
-	lea	160(%rdx), %rax
-	vpxor	$T0, $ACC8, $ACC8
-	vpand	32*1(%rdx), $B, $T0
-	vpxor	$Y, $ACC0, $ACC0
-	vpand	32*2(%rdx), $B, $Y
-	vpxor	$T0, $ACC1, $ACC1
-	vpand	32*3(%rdx), $B, $T0
-	vpxor	$Y, $ACC2, $ACC2
-	vpand	32*4-160(%rax), $B, $Y
-	vpxor	$T0, $ACC3, $ACC3
-	vpand	32*5-160(%rax), $B, $T0
-	vpxor	$Y, $ACC4, $ACC4
-	vpand	32*6-160(%rax), $B, $Y
-	vpxor	$T0, $ACC5, $ACC5
-	vpand	32*7-160(%rax), $B, $T0
-	vpxor	$Y, $ACC6, $ACC6
-	vpand	32*8-160(%rax), $B, $Y
-	vpxor	$T0, $ACC7, $ACC7
-	vpxor	$Y, $ACC8, $ACC8
-	`&STORE`
-
-	ret
-.size	avx2_select_n_store,.-avx2_select_n_store
-___
-$code.=<<___	if (0);				# inlined
-################################################################################
-# void avx2_mul_by2_x4(void* RESULTx4, void *Ax4);
-.type	avx2_mul_by2_x4,\@abi-omnipotent
-.align	32
-avx2_mul_by2_x4:
-	vmovdqa	32*0($a_ptr), $ACC0
-	lea	160($a_ptr), %rax
-	vmovdqa	32*1($a_ptr), $ACC1
-	vmovdqa	32*2($a_ptr), $ACC2
-	vmovdqa	32*3($a_ptr), $ACC3
-	vmovdqa	32*4-160(%rax), $ACC4
-	vmovdqa	32*5-160(%rax), $ACC5
-	vmovdqa	32*6-160(%rax), $ACC6
-	vmovdqa	32*7-160(%rax), $ACC7
-	vmovdqa	32*8-160(%rax), $ACC8
-
-	vpaddq	$ACC0, $ACC0, $ACC0
-	vpaddq	$ACC1, $ACC1, $ACC1
-	vpaddq	$ACC2, $ACC2, $ACC2
-	vpaddq	$ACC3, $ACC3, $ACC3
-	vpaddq	$ACC4, $ACC4, $ACC4
-	vpaddq	$ACC5, $ACC5, $ACC5
-	vpaddq	$ACC6, $ACC6, $ACC6
-	vpaddq	$ACC7, $ACC7, $ACC7
-	vpaddq	$ACC8, $ACC8, $ACC8
-
-	ret
-.size	avx2_mul_by2_x4,.-avx2_mul_by2_x4
-___
-my ($r_ptr_in,$a_ptr_in,$b_ptr_in)=("%rdi","%rsi","%rdx");
-my ($r_ptr,$a_ptr,$b_ptr)=("%r8","%r9","%r10");
-
-$code.=<<___;
-################################################################################
-# void ecp_nistz256_avx2_point_add_affine_x4(void* RESULTx4, void *Ax4, void *Bx4);
-.globl	ecp_nistz256_avx2_point_add_affine_x4
-.type	ecp_nistz256_avx2_point_add_affine_x4,\@function,3
-.align	32
-ecp_nistz256_avx2_point_add_affine_x4:
-	mov	%rsp, %rax
-	push    %rbp
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	lea	-8(%rax), %rbp
-
-# Result + 32*0 = Result.X
-# Result + 32*9 = Result.Y
-# Result + 32*18 = Result.Z
-
-# A + 32*0 = A.X
-# A + 32*9 = A.Y
-# A + 32*18 = A.Z
-
-# B + 32*0 = B.X
-# B + 32*9 = B.Y
-
-	sub	\$`32*9*8+32*2+32*8`, %rsp
-	and	\$-64, %rsp
-
-	mov	$r_ptr_in, $r_ptr
-	mov	$a_ptr_in, $a_ptr
-	mov	$b_ptr_in, $b_ptr
-
-	vmovdqa	32*0($a_ptr_in), %ymm0
-	vmovdqa	.LAVX2_AND_MASK(%rip), $AND_MASK
-	vpxor	%ymm1, %ymm1, %ymm1
-	lea	256($a_ptr_in), %rax		# size optimization
-	vpor	32*1($a_ptr_in), %ymm0, %ymm0
-	vpor	32*2($a_ptr_in), %ymm0, %ymm0
-	vpor	32*3($a_ptr_in), %ymm0, %ymm0
-	vpor	32*4-256(%rax), %ymm0, %ymm0
-	lea	256(%rax), %rcx			# size optimization
-	vpor	32*5-256(%rax), %ymm0, %ymm0
-	vpor	32*6-256(%rax), %ymm0, %ymm0
-	vpor	32*7-256(%rax), %ymm0, %ymm0
-	vpor	32*8-256(%rax), %ymm0, %ymm0
-	vpor	32*9-256(%rax), %ymm0, %ymm0
-	vpor	32*10-256(%rax), %ymm0, %ymm0
-	vpor	32*11-256(%rax), %ymm0, %ymm0
-	vpor	32*12-512(%rcx), %ymm0, %ymm0
-	vpor	32*13-512(%rcx), %ymm0, %ymm0
-	vpor	32*14-512(%rcx), %ymm0, %ymm0
-	vpor	32*15-512(%rcx), %ymm0, %ymm0
-	vpor	32*16-512(%rcx), %ymm0, %ymm0
-	vpor	32*17-512(%rcx), %ymm0, %ymm0
-	vpcmpeqq %ymm1, %ymm0, %ymm0
-	vmovdqa	%ymm0, `32*9*8`(%rsp)
-
-	vpxor	%ymm1, %ymm1, %ymm1
-	vmovdqa	32*0($b_ptr), %ymm0
-	lea	256($b_ptr), %rax		# size optimization
-	vpor	32*1($b_ptr), %ymm0, %ymm0
-	vpor	32*2($b_ptr), %ymm0, %ymm0
-	vpor	32*3($b_ptr), %ymm0, %ymm0
-	vpor	32*4-256(%rax), %ymm0, %ymm0
-	lea	256(%rax), %rcx			# size optimization
-	vpor	32*5-256(%rax), %ymm0, %ymm0
-	vpor	32*6-256(%rax), %ymm0, %ymm0
-	vpor	32*7-256(%rax), %ymm0, %ymm0
-	vpor	32*8-256(%rax), %ymm0, %ymm0
-	vpor	32*9-256(%rax), %ymm0, %ymm0
-	vpor	32*10-256(%rax), %ymm0, %ymm0
-	vpor	32*11-256(%rax), %ymm0, %ymm0
-	vpor	32*12-512(%rcx), %ymm0, %ymm0
-	vpor	32*13-512(%rcx), %ymm0, %ymm0
-	vpor	32*14-512(%rcx), %ymm0, %ymm0
-	vpor	32*15-512(%rcx), %ymm0, %ymm0
-	vpor	32*16-512(%rcx), %ymm0, %ymm0
-	vpor	32*17-512(%rcx), %ymm0, %ymm0
-	vpcmpeqq %ymm1, %ymm0, %ymm0
-	vmovdqa	%ymm0, `32*9*8+32`(%rsp)
-
-	#	Z1^2 = Z1*Z1
-	lea	`32*9*2`($a_ptr), %rsi
-	lea	`32*9*2`(%rsp), %rdi
-	lea	`32*9*8+32*2`(%rsp), %rcx	# temporary vector
-	call	avx2_sqr_x4
-	call	avx2_normalize_n_store
-
-	#	U2 = X2*Z1^2
-	lea	`32*9*0`($b_ptr), %rsi
-	lea	`32*9*2`(%rsp), %rdx
-	lea	`32*9*0`(%rsp), %rdi
-	call	avx2_mul_x4
-	#call	avx2_normalize
-	`&STORE`
-
-	#	S2 = Z1*Z1^2 = Z1^3
-	lea	`32*9*2`($a_ptr), %rsi
-	lea	`32*9*2`(%rsp), %rdx
-	lea	`32*9*1`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#	S2 = S2*Y2 = Y2*Z1^3
-	lea	`32*9*1`($b_ptr), %rsi
-	lea	`32*9*1`(%rsp), %rdx
-	lea	`32*9*1`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#	H = U2 - U1 = U2 - X1
-	lea	`32*9*0`(%rsp), %rsi
-	lea	`32*9*0`($a_ptr), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#	R = S2 - S1 = S2 - Y1
-	lea	`32*9*1`(%rsp), %rsi
-	lea	`32*9*1`($a_ptr), %rdx
-	lea	`32*9*4`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#	Z3 = H*Z1*Z2
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*2`($a_ptr), %rdx
-	lea	`32*9*2`($r_ptr), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize
-
-	lea	.LONE(%rip), %rsi
-	lea	`32*9*2`($a_ptr), %rdx
-	call	avx2_select_n_store
-
-	#	R^2 = R^2
-	lea	`32*9*4`(%rsp), %rsi
-	lea	`32*9*6`(%rsp), %rdi
-	lea	`32*9*8+32*2`(%rsp), %rcx	# temporary vector
-	call	avx2_sqr_x4
-	call	avx2_normalize_n_store
-
-	#	H^2 = H^2
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*5`(%rsp), %rdi
-	call	avx2_sqr_x4
-	call	avx2_normalize_n_store
-
-	#	H^3 = H^2*H
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*5`(%rsp), %rdx
-	lea	`32*9*7`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#	U2 = U1*H^2
-	lea	`32*9*0`($a_ptr), %rsi
-	lea	`32*9*5`(%rsp), %rdx
-	lea	`32*9*0`(%rsp), %rdi
-	call	avx2_mul_x4
-	#call	avx2_normalize
-	`&STORE`
-
-	#	Hsqr = U2*2
-	#lea	32*9*0(%rsp), %rsi
-	#lea	32*9*5(%rsp), %rdi
-	#call	avx2_mul_by2_x4
-
-	vpaddq	$ACC0, $ACC0, $ACC0	# inlined avx2_mul_by2_x4
-	lea	`32*9*5`(%rsp), %rdi
-	vpaddq	$ACC1, $ACC1, $ACC1
-	vpaddq	$ACC2, $ACC2, $ACC2
-	vpaddq	$ACC3, $ACC3, $ACC3
-	vpaddq	$ACC4, $ACC4, $ACC4
-	vpaddq	$ACC5, $ACC5, $ACC5
-	vpaddq	$ACC6, $ACC6, $ACC6
-	vpaddq	$ACC7, $ACC7, $ACC7
-	vpaddq	$ACC8, $ACC8, $ACC8
-	call	avx2_normalize_n_store
-
-	#	X3 = R^2 - H^3
-	#lea	32*9*6(%rsp), %rsi
-	#lea	32*9*7(%rsp), %rdx
-	#lea	32*9*5(%rsp), %rcx
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_sub_x4
-	#NORMALIZE
-	#STORE
-
-	#	X3 = X3 - U2*2
-	#lea	32*9*0($r_ptr), %rsi
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_sub_x4
-	#NORMALIZE
-	#STORE
-
-	lea	`32*9*6+128`(%rsp), %rsi
-	lea	.LAVX2_POLY_x2+128(%rip), %rax
-	lea	`32*9*7+128`(%rsp), %rdx
-	lea	`32*9*5+128`(%rsp), %rcx
-	lea	`32*9*0`($r_ptr), %rdi
-
-	vmovdqa	32*0-128(%rsi), $ACC0
-	vmovdqa	32*1-128(%rsi), $ACC1
-	vmovdqa	32*2-128(%rsi), $ACC2
-	vmovdqa	32*3-128(%rsi), $ACC3
-	vmovdqa	32*4-128(%rsi), $ACC4
-	vmovdqa	32*5-128(%rsi), $ACC5
-	vmovdqa	32*6-128(%rsi), $ACC6
-	vmovdqa	32*7-128(%rsi), $ACC7
-	vmovdqa	32*8-128(%rsi), $ACC8
-
-	vpaddq	32*0-128(%rax), $ACC0, $ACC0
-	vpaddq	32*1-128(%rax), $ACC1, $ACC1
-	vpaddq	32*2-128(%rax), $ACC2, $ACC2
-	vpaddq	32*3-128(%rax), $ACC3, $ACC3
-	vpaddq	32*4-128(%rax), $ACC4, $ACC4
-	vpaddq	32*5-128(%rax), $ACC5, $ACC5
-	vpaddq	32*6-128(%rax), $ACC6, $ACC6
-	vpaddq	32*7-128(%rax), $ACC7, $ACC7
-	vpaddq	32*8-128(%rax), $ACC8, $ACC8
-
-	vpsubq	32*0-128(%rdx), $ACC0, $ACC0
-	vpsubq	32*1-128(%rdx), $ACC1, $ACC1
-	vpsubq	32*2-128(%rdx), $ACC2, $ACC2
-	vpsubq	32*3-128(%rdx), $ACC3, $ACC3
-	vpsubq	32*4-128(%rdx), $ACC4, $ACC4
-	vpsubq	32*5-128(%rdx), $ACC5, $ACC5
-	vpsubq	32*6-128(%rdx), $ACC6, $ACC6
-	vpsubq	32*7-128(%rdx), $ACC7, $ACC7
-	vpsubq	32*8-128(%rdx), $ACC8, $ACC8
-
-	vpsubq	32*0-128(%rcx), $ACC0, $ACC0
-	vpsubq	32*1-128(%rcx), $ACC1, $ACC1
-	vpsubq	32*2-128(%rcx), $ACC2, $ACC2
-	vpsubq	32*3-128(%rcx), $ACC3, $ACC3
-	vpsubq	32*4-128(%rcx), $ACC4, $ACC4
-	vpsubq	32*5-128(%rcx), $ACC5, $ACC5
-	vpsubq	32*6-128(%rcx), $ACC6, $ACC6
-	vpsubq	32*7-128(%rcx), $ACC7, $ACC7
-	vpsubq	32*8-128(%rcx), $ACC8, $ACC8
-	call	avx2_normalize
-
-	lea	32*0($b_ptr), %rsi
-	lea	32*0($a_ptr), %rdx
-	call	avx2_select_n_store
-
-	#	H = U2 - X3
-	lea	`32*9*0`(%rsp), %rsi
-	lea	`32*9*0`($r_ptr), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*4`(%rsp), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#
-	lea	`32*9*7`(%rsp), %rsi
-	lea	`32*9*1`($a_ptr), %rdx
-	lea	`32*9*1`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*1`(%rsp), %rdx
-	lea	`32*9*1`($r_ptr), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize
-
-	lea	32*9($b_ptr), %rsi
-	lea	32*9($a_ptr), %rdx
-	call	avx2_select_n_store
-
-	#lea	32*9*0($r_ptr), %rsi
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_mul_by1_x4
-	#NORMALIZE
-	#STORE
-
-	lea	`32*9*1`($r_ptr), %rsi
-	lea	`32*9*1`($r_ptr), %rdi
-	call	avx2_mul_by1_x4
-	call	avx2_normalize_n_store
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	%xmm6, -16*10(%rbp)
-	movaps	%xmm7, -16*9(%rbp)
-	movaps	%xmm8, -16*8(%rbp)
-	movaps	%xmm9, -16*7(%rbp)
-	movaps	%xmm10, -16*6(%rbp)
-	movaps	%xmm11, -16*5(%rbp)
-	movaps	%xmm12, -16*4(%rbp)
-	movaps	%xmm13, -16*3(%rbp)
-	movaps	%xmm14, -16*2(%rbp)
-	movaps	%xmm15, -16*1(%rbp)
-___
-$code.=<<___;
-	mov	%rbp, %rsp
-	pop	%rbp
-	ret
-.size	ecp_nistz256_avx2_point_add_affine_x4,.-ecp_nistz256_avx2_point_add_affine_x4
-
-################################################################################
-# void ecp_nistz256_avx2_point_add_affines_x4(void* RESULTx4, void *Ax4, void *Bx4);
-.globl	ecp_nistz256_avx2_point_add_affines_x4
-.type	ecp_nistz256_avx2_point_add_affines_x4,\@function,3
-.align	32
-ecp_nistz256_avx2_point_add_affines_x4:
-	mov	%rsp, %rax
-	push    %rbp
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	lea	-8(%rax), %rbp
-
-# Result + 32*0 = Result.X
-# Result + 32*9 = Result.Y
-# Result + 32*18 = Result.Z
-
-# A + 32*0 = A.X
-# A + 32*9 = A.Y
-
-# B + 32*0 = B.X
-# B + 32*9 = B.Y
-
-	sub	\$`32*9*8+32*2+32*8`, %rsp
-	and	\$-64, %rsp
-
-	mov	$r_ptr_in, $r_ptr
-	mov	$a_ptr_in, $a_ptr
-	mov	$b_ptr_in, $b_ptr
-
-	vmovdqa	32*0($a_ptr_in), %ymm0
-	vmovdqa	.LAVX2_AND_MASK(%rip), $AND_MASK
-	vpxor	%ymm1, %ymm1, %ymm1
-	lea	256($a_ptr_in), %rax		# size optimization
-	vpor	32*1($a_ptr_in), %ymm0, %ymm0
-	vpor	32*2($a_ptr_in), %ymm0, %ymm0
-	vpor	32*3($a_ptr_in), %ymm0, %ymm0
-	vpor	32*4-256(%rax), %ymm0, %ymm0
-	lea	256(%rax), %rcx			# size optimization
-	vpor	32*5-256(%rax), %ymm0, %ymm0
-	vpor	32*6-256(%rax), %ymm0, %ymm0
-	vpor	32*7-256(%rax), %ymm0, %ymm0
-	vpor	32*8-256(%rax), %ymm0, %ymm0
-	vpor	32*9-256(%rax), %ymm0, %ymm0
-	vpor	32*10-256(%rax), %ymm0, %ymm0
-	vpor	32*11-256(%rax), %ymm0, %ymm0
-	vpor	32*12-512(%rcx), %ymm0, %ymm0
-	vpor	32*13-512(%rcx), %ymm0, %ymm0
-	vpor	32*14-512(%rcx), %ymm0, %ymm0
-	vpor	32*15-512(%rcx), %ymm0, %ymm0
-	vpor	32*16-512(%rcx), %ymm0, %ymm0
-	vpor	32*17-512(%rcx), %ymm0, %ymm0
-	vpcmpeqq %ymm1, %ymm0, %ymm0
-	vmovdqa	%ymm0, `32*9*8`(%rsp)
-
-	vpxor	%ymm1, %ymm1, %ymm1
-	vmovdqa	32*0($b_ptr), %ymm0
-	lea	256($b_ptr), %rax		# size optimization
-	vpor	32*1($b_ptr), %ymm0, %ymm0
-	vpor	32*2($b_ptr), %ymm0, %ymm0
-	vpor	32*3($b_ptr), %ymm0, %ymm0
-	vpor	32*4-256(%rax), %ymm0, %ymm0
-	lea	256(%rax), %rcx			# size optimization
-	vpor	32*5-256(%rax), %ymm0, %ymm0
-	vpor	32*6-256(%rax), %ymm0, %ymm0
-	vpor	32*7-256(%rax), %ymm0, %ymm0
-	vpor	32*8-256(%rax), %ymm0, %ymm0
-	vpor	32*9-256(%rax), %ymm0, %ymm0
-	vpor	32*10-256(%rax), %ymm0, %ymm0
-	vpor	32*11-256(%rax), %ymm0, %ymm0
-	vpor	32*12-512(%rcx), %ymm0, %ymm0
-	vpor	32*13-512(%rcx), %ymm0, %ymm0
-	vpor	32*14-512(%rcx), %ymm0, %ymm0
-	vpor	32*15-512(%rcx), %ymm0, %ymm0
-	vpor	32*16-512(%rcx), %ymm0, %ymm0
-	vpor	32*17-512(%rcx), %ymm0, %ymm0
-	vpcmpeqq %ymm1, %ymm0, %ymm0
-	vmovdqa	%ymm0, `32*9*8+32`(%rsp)
-
-	#	H = U2 - U1 = X2 - X1
-	lea	`32*9*0`($b_ptr), %rsi
-	lea	`32*9*0`($a_ptr), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#	R = S2 - S1 = Y2 - Y1
-	lea	`32*9*1`($b_ptr), %rsi
-	lea	`32*9*1`($a_ptr), %rdx
-	lea	`32*9*4`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#	Z3 = H*Z1*Z2 = H
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*2`($r_ptr), %rdi
-	call	avx2_mul_by1_x4
-	call	avx2_normalize
-
-	vmovdqa	`32*9*8`(%rsp), $B
-	vpor	`32*9*8+32`(%rsp), $B, $B
-
-	vpandn	$ACC0, $B, $ACC0
-	lea	.LONE+128(%rip), %rax
-	vpandn	$ACC1, $B, $ACC1
-	vpandn	$ACC2, $B, $ACC2
-	vpandn	$ACC3, $B, $ACC3
-	vpandn	$ACC4, $B, $ACC4
-	vpandn	$ACC5, $B, $ACC5
-	vpandn	$ACC6, $B, $ACC6
-	vpandn	$ACC7, $B, $ACC7
-
-	vpand	32*0-128(%rax), $B, $T0
-	 vpandn	$ACC8, $B, $ACC8
-	vpand	32*1-128(%rax), $B, $Y
-	vpxor	$T0, $ACC0, $ACC0
-	vpand	32*2-128(%rax), $B, $T0
-	vpxor	$Y, $ACC1, $ACC1
-	vpand	32*3-128(%rax), $B, $Y
-	vpxor	$T0, $ACC2, $ACC2
-	vpand	32*4-128(%rax), $B, $T0
-	vpxor	$Y, $ACC3, $ACC3
-	vpand	32*5-128(%rax), $B, $Y
-	vpxor	$T0, $ACC4, $ACC4
-	vpand	32*6-128(%rax), $B, $T0
-	vpxor	$Y, $ACC5, $ACC5
-	vpand	32*7-128(%rax), $B, $Y
-	vpxor	$T0, $ACC6, $ACC6
-	vpand	32*8-128(%rax), $B, $T0
-	vpxor	$Y, $ACC7, $ACC7
-	vpxor	$T0, $ACC8, $ACC8
-	`&STORE`
-
-	#	R^2 = R^2
-	lea	`32*9*4`(%rsp), %rsi
-	lea	`32*9*6`(%rsp), %rdi
-	lea	`32*9*8+32*2`(%rsp), %rcx	# temporary vector
-	call	avx2_sqr_x4
-	call	avx2_normalize_n_store
-
-	#	H^2 = H^2
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*5`(%rsp), %rdi
-	call	avx2_sqr_x4
-	call	avx2_normalize_n_store
-
-	#	H^3 = H^2*H
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*5`(%rsp), %rdx
-	lea	`32*9*7`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#	U2 = U1*H^2
-	lea	`32*9*0`($a_ptr), %rsi
-	lea	`32*9*5`(%rsp), %rdx
-	lea	`32*9*0`(%rsp), %rdi
-	call	avx2_mul_x4
-	#call	avx2_normalize
-	`&STORE`
-
-	#	Hsqr = U2*2
-	#lea	32*9*0(%rsp), %rsi
-	#lea	32*9*5(%rsp), %rdi
-	#call	avx2_mul_by2_x4
-
-	vpaddq	$ACC0, $ACC0, $ACC0	# inlined avx2_mul_by2_x4
-	lea	`32*9*5`(%rsp), %rdi
-	vpaddq	$ACC1, $ACC1, $ACC1
-	vpaddq	$ACC2, $ACC2, $ACC2
-	vpaddq	$ACC3, $ACC3, $ACC3
-	vpaddq	$ACC4, $ACC4, $ACC4
-	vpaddq	$ACC5, $ACC5, $ACC5
-	vpaddq	$ACC6, $ACC6, $ACC6
-	vpaddq	$ACC7, $ACC7, $ACC7
-	vpaddq	$ACC8, $ACC8, $ACC8
-	call	avx2_normalize_n_store
-
-	#	X3 = R^2 - H^3
-	#lea	32*9*6(%rsp), %rsi
-	#lea	32*9*7(%rsp), %rdx
-	#lea	32*9*5(%rsp), %rcx
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_sub_x4
-	#NORMALIZE
-	#STORE
-
-	#	X3 = X3 - U2*2
-	#lea	32*9*0($r_ptr), %rsi
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_sub_x4
-	#NORMALIZE
-	#STORE
-
-	lea	`32*9*6+128`(%rsp), %rsi
-	lea	.LAVX2_POLY_x2+128(%rip), %rax
-	lea	`32*9*7+128`(%rsp), %rdx
-	lea	`32*9*5+128`(%rsp), %rcx
-	lea	`32*9*0`($r_ptr), %rdi
-
-	vmovdqa	32*0-128(%rsi), $ACC0
-	vmovdqa	32*1-128(%rsi), $ACC1
-	vmovdqa	32*2-128(%rsi), $ACC2
-	vmovdqa	32*3-128(%rsi), $ACC3
-	vmovdqa	32*4-128(%rsi), $ACC4
-	vmovdqa	32*5-128(%rsi), $ACC5
-	vmovdqa	32*6-128(%rsi), $ACC6
-	vmovdqa	32*7-128(%rsi), $ACC7
-	vmovdqa	32*8-128(%rsi), $ACC8
-
-	vpaddq	32*0-128(%rax), $ACC0, $ACC0
-	vpaddq	32*1-128(%rax), $ACC1, $ACC1
-	vpaddq	32*2-128(%rax), $ACC2, $ACC2
-	vpaddq	32*3-128(%rax), $ACC3, $ACC3
-	vpaddq	32*4-128(%rax), $ACC4, $ACC4
-	vpaddq	32*5-128(%rax), $ACC5, $ACC5
-	vpaddq	32*6-128(%rax), $ACC6, $ACC6
-	vpaddq	32*7-128(%rax), $ACC7, $ACC7
-	vpaddq	32*8-128(%rax), $ACC8, $ACC8
-
-	vpsubq	32*0-128(%rdx), $ACC0, $ACC0
-	vpsubq	32*1-128(%rdx), $ACC1, $ACC1
-	vpsubq	32*2-128(%rdx), $ACC2, $ACC2
-	vpsubq	32*3-128(%rdx), $ACC3, $ACC3
-	vpsubq	32*4-128(%rdx), $ACC4, $ACC4
-	vpsubq	32*5-128(%rdx), $ACC5, $ACC5
-	vpsubq	32*6-128(%rdx), $ACC6, $ACC6
-	vpsubq	32*7-128(%rdx), $ACC7, $ACC7
-	vpsubq	32*8-128(%rdx), $ACC8, $ACC8
-
-	vpsubq	32*0-128(%rcx), $ACC0, $ACC0
-	vpsubq	32*1-128(%rcx), $ACC1, $ACC1
-	vpsubq	32*2-128(%rcx), $ACC2, $ACC2
-	vpsubq	32*3-128(%rcx), $ACC3, $ACC3
-	vpsubq	32*4-128(%rcx), $ACC4, $ACC4
-	vpsubq	32*5-128(%rcx), $ACC5, $ACC5
-	vpsubq	32*6-128(%rcx), $ACC6, $ACC6
-	vpsubq	32*7-128(%rcx), $ACC7, $ACC7
-	vpsubq	32*8-128(%rcx), $ACC8, $ACC8
-	call	avx2_normalize
-
-	lea	32*0($b_ptr), %rsi
-	lea	32*0($a_ptr), %rdx
-	call	avx2_select_n_store
-
-	#	H = U2 - X3
-	lea	`32*9*0`(%rsp), %rsi
-	lea	`32*9*0`($r_ptr), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize_n_store
-
-	#	H = H*R
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*4`(%rsp), %rdx
-	lea	`32*9*3`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#	S2 = S1 * H^3
-	lea	`32*9*7`(%rsp), %rsi
-	lea	`32*9*1`($a_ptr), %rdx
-	lea	`32*9*1`(%rsp), %rdi
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	#
-	lea	`32*9*3`(%rsp), %rsi
-	lea	`32*9*1`(%rsp), %rdx
-	lea	`32*9*1`($r_ptr), %rdi
-	call	avx2_sub_x4
-	call	avx2_normalize
-
-	lea	32*9($b_ptr), %rsi
-	lea	32*9($a_ptr), %rdx
-	call	avx2_select_n_store
-
-	#lea	32*9*0($r_ptr), %rsi
-	#lea	32*9*0($r_ptr), %rdi
-	#call	avx2_mul_by1_x4
-	#NORMALIZE
-	#STORE
-
-	lea	`32*9*1`($r_ptr), %rsi
-	lea	`32*9*1`($r_ptr), %rdi
-	call	avx2_mul_by1_x4
-	call	avx2_normalize_n_store
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	%xmm6, -16*10(%rbp)
-	movaps	%xmm7, -16*9(%rbp)
-	movaps	%xmm8, -16*8(%rbp)
-	movaps	%xmm9, -16*7(%rbp)
-	movaps	%xmm10, -16*6(%rbp)
-	movaps	%xmm11, -16*5(%rbp)
-	movaps	%xmm12, -16*4(%rbp)
-	movaps	%xmm13, -16*3(%rbp)
-	movaps	%xmm14, -16*2(%rbp)
-	movaps	%xmm15, -16*1(%rbp)
-___
-$code.=<<___;
-	mov	%rbp, %rsp
-	pop	%rbp
-	ret
-.size	ecp_nistz256_avx2_point_add_affines_x4,.-ecp_nistz256_avx2_point_add_affines_x4
-
-################################################################################
-# void ecp_nistz256_avx2_to_mont(void* RESULTx4, void *Ax4);
-.globl	ecp_nistz256_avx2_to_mont
-.type	ecp_nistz256_avx2_to_mont,\@function,2
-.align	32
-ecp_nistz256_avx2_to_mont:
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-8-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	vmovdqa	.LAVX2_AND_MASK(%rip), $AND_MASK
-	lea	.LTO_MONT_AVX2(%rip), %rdx
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	16*0(%rsp), %xmm6
-	movaps	16*1(%rsp), %xmm7
-	movaps	16*2(%rsp), %xmm8
-	movaps	16*3(%rsp), %xmm9
-	movaps	16*4(%rsp), %xmm10
-	movaps	16*5(%rsp), %xmm11
-	movaps	16*6(%rsp), %xmm12
-	movaps	16*7(%rsp), %xmm13
-	movaps	16*8(%rsp), %xmm14
-	movaps	16*9(%rsp), %xmm15
-	lea	8+16*10(%rsp), %rsp
-___
-$code.=<<___;
-	ret
-.size	ecp_nistz256_avx2_to_mont,.-ecp_nistz256_avx2_to_mont
-
-################################################################################
-# void ecp_nistz256_avx2_from_mont(void* RESULTx4, void *Ax4);
-.globl	ecp_nistz256_avx2_from_mont
-.type	ecp_nistz256_avx2_from_mont,\@function,2
-.align	32
-ecp_nistz256_avx2_from_mont:
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-8-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	vmovdqa	.LAVX2_AND_MASK(%rip), $AND_MASK
-	lea	.LFROM_MONT_AVX2(%rip), %rdx
-	call	avx2_mul_x4
-	call	avx2_normalize_n_store
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	16*0(%rsp), %xmm6
-	movaps	16*1(%rsp), %xmm7
-	movaps	16*2(%rsp), %xmm8
-	movaps	16*3(%rsp), %xmm9
-	movaps	16*4(%rsp), %xmm10
-	movaps	16*5(%rsp), %xmm11
-	movaps	16*6(%rsp), %xmm12
-	movaps	16*7(%rsp), %xmm13
-	movaps	16*8(%rsp), %xmm14
-	movaps	16*9(%rsp), %xmm15
-	lea	8+16*10(%rsp), %rsp
-___
-$code.=<<___;
-	ret
-.size	ecp_nistz256_avx2_from_mont,.-ecp_nistz256_avx2_from_mont
-
-################################################################################
-# void ecp_nistz256_avx2_set1(void* RESULTx4);
-.globl	ecp_nistz256_avx2_set1
-.type	ecp_nistz256_avx2_set1,\@function,1
-.align	32
-ecp_nistz256_avx2_set1:
-	lea	.LONE+128(%rip), %rax
-	lea	128(%rdi), %rdi
-	vzeroupper
-	vmovdqa	32*0-128(%rax), %ymm0
-	vmovdqa	32*1-128(%rax), %ymm1
-	vmovdqa	32*2-128(%rax), %ymm2
-	vmovdqa	32*3-128(%rax), %ymm3
-	vmovdqa	32*4-128(%rax), %ymm4
-	vmovdqa	32*5-128(%rax), %ymm5
-	vmovdqa	%ymm0, 32*0-128(%rdi)
-	vmovdqa	32*6-128(%rax), %ymm0
-	vmovdqa	%ymm1, 32*1-128(%rdi)
-	vmovdqa	32*7-128(%rax), %ymm1
-	vmovdqa	%ymm2, 32*2-128(%rdi)
-	vmovdqa	32*8-128(%rax), %ymm2
-	vmovdqa	%ymm3, 32*3-128(%rdi)
-	vmovdqa	%ymm4, 32*4-128(%rdi)
-	vmovdqa	%ymm5, 32*5-128(%rdi)
-	vmovdqa	%ymm0, 32*6-128(%rdi)
-	vmovdqa	%ymm1, 32*7-128(%rdi)
-	vmovdqa	%ymm2, 32*8-128(%rdi)
-
-	vzeroupper
-	ret
-.size	ecp_nistz256_avx2_set1,.-ecp_nistz256_avx2_set1
-___
-}
-{
-################################################################################
-# void ecp_nistz256_avx2_multi_gather_w7(void* RESULT, void *in,
-#			    int index0, int index1, int index2, int index3);
-################################################################################
-
-my ($val,$in_t,$index0,$index1,$index2,$index3)=("%rdi","%rsi","%edx","%ecx","%r8d","%r9d");
-my ($INDEX0,$INDEX1,$INDEX2,$INDEX3)=map("%ymm$_",(0..3));
-my ($R0a,$R0b,$R1a,$R1b,$R2a,$R2b,$R3a,$R3b)=map("%ymm$_",(4..11));
-my ($M0,$T0,$T1,$TMP0)=map("%ymm$_",(12..15));
-
-$code.=<<___;
-.globl	ecp_nistz256_avx2_multi_gather_w7
-.type	ecp_nistz256_avx2_multi_gather_w7,\@function,6
-.align	32
-ecp_nistz256_avx2_multi_gather_w7:
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	lea	-8-16*10(%rsp), %rsp
-	vmovaps	%xmm6, -8-16*10(%rax)
-	vmovaps	%xmm7, -8-16*9(%rax)
-	vmovaps	%xmm8, -8-16*8(%rax)
-	vmovaps	%xmm9, -8-16*7(%rax)
-	vmovaps	%xmm10, -8-16*6(%rax)
-	vmovaps	%xmm11, -8-16*5(%rax)
-	vmovaps	%xmm12, -8-16*4(%rax)
-	vmovaps	%xmm13, -8-16*3(%rax)
-	vmovaps	%xmm14, -8-16*2(%rax)
-	vmovaps	%xmm15, -8-16*1(%rax)
-___
-$code.=<<___;
-	lea	.LIntOne(%rip), %rax
-
-	vmovd	$index0, %xmm0
-	vmovd	$index1, %xmm1
-	vmovd	$index2, %xmm2
-	vmovd	$index3, %xmm3
-
-	vpxor	$R0a, $R0a, $R0a
-	vpxor	$R0b, $R0b, $R0b
-	vpxor	$R1a, $R1a, $R1a
-	vpxor	$R1b, $R1b, $R1b
-	vpxor	$R2a, $R2a, $R2a
-	vpxor	$R2b, $R2b, $R2b
-	vpxor	$R3a, $R3a, $R3a
-	vpxor	$R3b, $R3b, $R3b
-	vmovdqa	(%rax), $M0
-
-	vpermd	$INDEX0, $R0a, $INDEX0
-	vpermd	$INDEX1, $R0a, $INDEX1
-	vpermd	$INDEX2, $R0a, $INDEX2
-	vpermd	$INDEX3, $R0a, $INDEX3
-
-	mov	\$64, %ecx
-	lea	112($val), $val		# size optimization
-	jmp	.Lmulti_select_loop_avx2
-
-# INDEX=0, corresponds to the point at infty (0,0)
-.align	32
-.Lmulti_select_loop_avx2:
-	vpcmpeqd	$INDEX0, $M0, $TMP0
-
-	vmovdqa		`32*0+32*64*2*0`($in_t), $T0
-	vmovdqa		`32*1+32*64*2*0`($in_t), $T1
-	vpand		$TMP0, $T0, $T0
-	vpand		$TMP0, $T1, $T1
-	vpxor		$T0, $R0a, $R0a
-	vpxor		$T1, $R0b, $R0b
-
-	vpcmpeqd	$INDEX1, $M0, $TMP0
-
-	vmovdqa		`32*0+32*64*2*1`($in_t), $T0
-	vmovdqa		`32*1+32*64*2*1`($in_t), $T1
-	vpand		$TMP0, $T0, $T0
-	vpand		$TMP0, $T1, $T1
-	vpxor		$T0, $R1a, $R1a
-	vpxor		$T1, $R1b, $R1b
-
-	vpcmpeqd	$INDEX2, $M0, $TMP0
-
-	vmovdqa		`32*0+32*64*2*2`($in_t), $T0
-	vmovdqa		`32*1+32*64*2*2`($in_t), $T1
-	vpand		$TMP0, $T0, $T0
-	vpand		$TMP0, $T1, $T1
-	vpxor		$T0, $R2a, $R2a
-	vpxor		$T1, $R2b, $R2b
-
-	vpcmpeqd	$INDEX3, $M0, $TMP0
-
-	vmovdqa		`32*0+32*64*2*3`($in_t), $T0
-	vmovdqa		`32*1+32*64*2*3`($in_t), $T1
-	vpand		$TMP0, $T0, $T0
-	vpand		$TMP0, $T1, $T1
-	vpxor		$T0, $R3a, $R3a
-	vpxor		$T1, $R3b, $R3b
-
-	vpaddd		(%rax), $M0, $M0	# increment
-	lea		32*2($in_t), $in_t
-
-        dec	%ecx
-	jnz	.Lmulti_select_loop_avx2
-
-	vmovdqu	$R0a, 32*0-112($val)
-	vmovdqu	$R0b, 32*1-112($val)
-	vmovdqu	$R1a, 32*2-112($val)
-	vmovdqu	$R1b, 32*3-112($val)
-	vmovdqu	$R2a, 32*4-112($val)
-	vmovdqu	$R2b, 32*5-112($val)
-	vmovdqu	$R3a, 32*6-112($val)
-	vmovdqu	$R3b, 32*7-112($val)
-
-	vzeroupper
-___
-$code.=<<___	if ($win64);
-	movaps	16*0(%rsp), %xmm6
-	movaps	16*1(%rsp), %xmm7
-	movaps	16*2(%rsp), %xmm8
-	movaps	16*3(%rsp), %xmm9
-	movaps	16*4(%rsp), %xmm10
-	movaps	16*5(%rsp), %xmm11
-	movaps	16*6(%rsp), %xmm12
-	movaps	16*7(%rsp), %xmm13
-	movaps	16*8(%rsp), %xmm14
-	movaps	16*9(%rsp), %xmm15
-	lea	8+16*10(%rsp), %rsp
-___
-$code.=<<___;
-	ret
-.size	ecp_nistz256_avx2_multi_gather_w7,.-ecp_nistz256_avx2_multi_gather_w7
-
-.extern	OPENSSL_ia32cap_P
-.globl	ecp_nistz_avx2_eligible
-.type	ecp_nistz_avx2_eligible,\@abi-omnipotent
-.align	32
-ecp_nistz_avx2_eligible:
-	mov	OPENSSL_ia32cap_P+8(%rip),%eax
-	shr	\$5,%eax
-	and	\$1,%eax
-	ret
-.size	ecp_nistz_avx2_eligible,.-ecp_nistz_avx2_eligible
-___
-}
-}} else {{	# assembler is too old
-$code.=<<___;
-.text
-
-.globl	ecp_nistz256_avx2_transpose_convert
-.globl	ecp_nistz256_avx2_convert_transpose_back
-.globl	ecp_nistz256_avx2_point_add_affine_x4
-.globl	ecp_nistz256_avx2_point_add_affines_x4
-.globl	ecp_nistz256_avx2_to_mont
-.globl	ecp_nistz256_avx2_from_mont
-.globl	ecp_nistz256_avx2_set1
-.globl	ecp_nistz256_avx2_multi_gather_w7
-.type	ecp_nistz256_avx2_multi_gather_w7,\@abi-omnipotent
-ecp_nistz256_avx2_transpose_convert:
-ecp_nistz256_avx2_convert_transpose_back:
-ecp_nistz256_avx2_point_add_affine_x4:
-ecp_nistz256_avx2_point_add_affines_x4:
-ecp_nistz256_avx2_to_mont:
-ecp_nistz256_avx2_from_mont:
-ecp_nistz256_avx2_set1:
-ecp_nistz256_avx2_multi_gather_w7:
-	.byte	0x0f,0x0b	# ud2
-	ret
-.size	ecp_nistz256_avx2_multi_gather_w7,.-ecp_nistz256_avx2_multi_gather_w7
-
-.globl	ecp_nistz_avx2_eligible
-.type	ecp_nistz_avx2_eligible,\@abi-omnipotent
-ecp_nistz_avx2_eligible:
-	xor	%eax,%eax
-	ret
-.size	ecp_nistz_avx2_eligible,.-ecp_nistz_avx2_eligible
-___
-}}
-
-foreach (split("\n",$code)) {
-	s/\`([^\`]*)\`/eval($1)/geo;
-
-	print $_,"\n";
-}
-
-close STDOUT or die "error closing STDOUT: $!";
diff --git a/crypto/ec/asm/ecp_nistz256-x86_64.pl b/crypto/ec/asm/ecp_nistz256-x86_64.pl
index de9b194510bf..b50ee70191b2 100755
--- a/crypto/ec/asm/ecp_nistz256-x86_64.pl
+++ b/crypto/ec/asm/ecp_nistz256-x86_64.pl
@@ -72,7 +72,7 @@ if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	$addx = ($1>=12);
 }
 
-if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
+if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
 	my $ver = $2 + $3/100.0;	# 3.1->3.01, 3.10->3.10
 	$avx = ($ver>=3.0) + ($ver>=3.01);
 	$addx = ($ver>=3.03);
diff --git a/crypto/ec/asm/x25519-x86_64.pl b/crypto/ec/asm/x25519-x86_64.pl
index 3d9d1dc1ad0c..62599dacaccd 100755
--- a/crypto/ec/asm/x25519-x86_64.pl
+++ b/crypto/ec/asm/x25519-x86_64.pl
@@ -90,7 +90,7 @@ if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	$addx = ($1>=12);
 }
 
-if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
+if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
 	my $ver = $2 + $3/100.0;	# 3.1->3.01, 3.10->3.10
 	$addx = ($ver>=3.03);
 }
diff --git a/crypto/ec/ec_ameth.c b/crypto/ec/ec_ameth.c
index 221038373921..5098bd7a6602 100644
--- a/crypto/ec/ec_ameth.c
+++ b/crypto/ec/ec_ameth.c
@@ -1,5 +1,5 @@
 /*
- * Copyright 2006-2019 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 2006-2020 The OpenSSL Project Authors. All Rights Reserved.
  *
  * Licensed under the OpenSSL license (the "License").  You may not use
  * this file except in compliance with the License.  You can obtain a copy
@@ -23,7 +23,7 @@ static int ecdh_cms_decrypt(CMS_RecipientInfo *ri);
 static int ecdh_cms_encrypt(CMS_RecipientInfo *ri);
 #endif
 
-static int eckey_param2type(int *pptype, void **ppval, EC_KEY *ec_key)
+static int eckey_param2type(int *pptype, void **ppval, const EC_KEY *ec_key)
 {
     const EC_GROUP *group;
     int nid;
@@ -35,7 +35,14 @@ static int eckey_param2type(int *pptype, void **ppval, EC_KEY *ec_key)
         && (nid = EC_GROUP_get_curve_name(group)))
         /* we have a 'named curve' => just set the OID */
     {
-        *ppval = OBJ_nid2obj(nid);
+        ASN1_OBJECT *asn1obj = OBJ_nid2obj(nid);
+
+        if (asn1obj == NULL || OBJ_length(asn1obj) == 0) {
+            ASN1_OBJECT_free(asn1obj);
+            ECerr(EC_F_ECKEY_PARAM2TYPE, EC_R_MISSING_OID);
+            return 0;
+        }
+        *ppval = asn1obj;
         *pptype = V_ASN1_OBJECT;
     } else {                    /* explicit parameters */
 
@@ -43,7 +50,17 @@ static int eckey_param2type(int *pptype, void **ppval, EC_KEY *ec_key)
         pstr = ASN1_STRING_new();
         if (pstr == NULL)
             return 0;
-        pstr->length = i2d_ECParameters(ec_key, &pstr->data);
+
+        /*
+         * The cast in the following line is intentional as the
+         * `i2d_ECParameters` signature can't be constified (see discussion at
+         * https://github.com/openssl/openssl/pull/9347 where related and
+         * required constification backports were rejected).
+         *
+         * This cast should be safe anyway, because we can expect
+         * `i2d_ECParameters()` to treat the first argument as if it was const.
+         */
+        pstr->length = i2d_ECParameters((EC_KEY *)ec_key, &pstr->data);
         if (pstr->length <= 0) {
             ASN1_STRING_free(pstr);
             ECerr(EC_F_ECKEY_PARAM2TYPE, ERR_R_EC_LIB);
@@ -57,7 +74,7 @@ static int eckey_param2type(int *pptype, void **ppval, EC_KEY *ec_key)
 
 static int eckey_pub_encode(X509_PUBKEY *pk, const EVP_PKEY *pkey)
 {
-    EC_KEY *ec_key = pkey->pkey.ec;
+    const EC_KEY *ec_key = pkey->pkey.ec;
     void *pval = NULL;
     int ptype;
     unsigned char *penc = NULL, *p;
diff --git a/crypto/ec/ec_asn1.c b/crypto/ec/ec_asn1.c
index 006f9a5dea17..7b7c75ce8443 100644
--- a/crypto/ec/ec_asn1.c
+++ b/crypto/ec/ec_asn1.c
@@ -137,6 +137,12 @@ struct ec_parameters_st {
     ASN1_INTEGER *cofactor;
 } /* ECPARAMETERS */ ;
 
+typedef enum {
+    ECPKPARAMETERS_TYPE_NAMED = 0,
+    ECPKPARAMETERS_TYPE_EXPLICIT,
+    ECPKPARAMETERS_TYPE_IMPLICIT
+} ecpk_parameters_type_t;
+
 struct ecpk_parameters_st {
     int type;
     union {
@@ -535,9 +541,10 @@ ECPKPARAMETERS *EC_GROUP_get_ecpkparameters(const EC_GROUP *group,
             return NULL;
         }
     } else {
-        if (ret->type == 0)
+        if (ret->type == ECPKPARAMETERS_TYPE_NAMED)
             ASN1_OBJECT_free(ret->value.named_curve);
-        else if (ret->type == 1 && ret->value.parameters)
+        else if (ret->type == ECPKPARAMETERS_TYPE_EXPLICIT
+                 && ret->value.parameters != NULL)
             ECPARAMETERS_free(ret->value.parameters);
     }
 
@@ -547,15 +554,22 @@ ECPKPARAMETERS *EC_GROUP_get_ecpkparameters(const EC_GROUP *group,
          */
         tmp = EC_GROUP_get_curve_name(group);
         if (tmp) {
-            ret->type = 0;
-            if ((ret->value.named_curve = OBJ_nid2obj(tmp)) == NULL)
+            ASN1_OBJECT *asn1obj = OBJ_nid2obj(tmp);
+
+            if (asn1obj == NULL || OBJ_length(asn1obj) == 0) {
+                ASN1_OBJECT_free(asn1obj);
+                ECerr(EC_F_EC_GROUP_GET_ECPKPARAMETERS, EC_R_MISSING_OID);
                 ok = 0;
+            } else {
+                ret->type = ECPKPARAMETERS_TYPE_NAMED;
+                ret->value.named_curve = asn1obj;
+            }
         } else
             /* we don't know the nid => ERROR */
             ok = 0;
     } else {
         /* use the ECPARAMETERS structure */
-        ret->type = 1;
+        ret->type = ECPKPARAMETERS_TYPE_EXPLICIT;
         if ((ret->value.parameters =
              EC_GROUP_get_ecparameters(group, NULL)) == NULL)
             ok = 0;
@@ -894,7 +908,8 @@ EC_GROUP *EC_GROUP_new_from_ecpkparameters(const ECPKPARAMETERS *params)
         return NULL;
     }
 
-    if (params->type == 0) {    /* the curve is given by an OID */
+    if (params->type == ECPKPARAMETERS_TYPE_NAMED) {
+        /* the curve is given by an OID */
         tmp = OBJ_obj2nid(params->value.named_curve);
         if ((ret = EC_GROUP_new_by_curve_name(tmp)) == NULL) {
             ECerr(EC_F_EC_GROUP_NEW_FROM_ECPKPARAMETERS,
@@ -902,15 +917,16 @@ EC_GROUP *EC_GROUP_new_from_ecpkparameters(const ECPKPARAMETERS *params)
             return NULL;
         }
         EC_GROUP_set_asn1_flag(ret, OPENSSL_EC_NAMED_CURVE);
-    } else if (params->type == 1) { /* the parameters are given by a
-                                     * ECPARAMETERS structure */
+    } else if (params->type == ECPKPARAMETERS_TYPE_EXPLICIT) {
+        /* the parameters are given by an ECPARAMETERS structure */
         ret = EC_GROUP_new_from_ecparameters(params->value.parameters);
         if (!ret) {
             ECerr(EC_F_EC_GROUP_NEW_FROM_ECPKPARAMETERS, ERR_R_EC_LIB);
             return NULL;
         }
         EC_GROUP_set_asn1_flag(ret, OPENSSL_EC_EXPLICIT_CURVE);
-    } else if (params->type == 2) { /* implicitlyCA */
+    } else if (params->type == ECPKPARAMETERS_TYPE_IMPLICIT) {
+        /* implicit parameters inherited from CA - unsupported */
         return NULL;
     } else {
         ECerr(EC_F_EC_GROUP_NEW_FROM_ECPKPARAMETERS, EC_R_ASN1_ERROR);
@@ -940,6 +956,9 @@ EC_GROUP *d2i_ECPKParameters(EC_GROUP **a, const unsigned char **in, long len)
         return NULL;
     }
 
+    if (params->type == ECPKPARAMETERS_TYPE_EXPLICIT)
+        group->decoded_from_explicit_params = 1;
+
     if (a) {
         EC_GROUP_free(*a);
         *a = group;
@@ -991,6 +1010,9 @@ EC_KEY *d2i_ECPrivateKey(EC_KEY **a, const unsigned char **in, long len)
     if (priv_key->parameters) {
         EC_GROUP_free(ret->group);
         ret->group = EC_GROUP_new_from_ecpkparameters(priv_key->parameters);
+        if (ret->group != NULL
+            && priv_key->parameters->type == ECPKPARAMETERS_TYPE_EXPLICIT)
+            ret->group->decoded_from_explicit_params = 1;
     }
 
     if (ret->group == NULL) {
diff --git a/crypto/ec/ec_err.c b/crypto/ec/ec_err.c
index ce3493823218..bfe74226503e 100644
--- a/crypto/ec/ec_err.c
+++ b/crypto/ec/ec_err.c
@@ -1,6 +1,6 @@
 /*
  * Generated by util/mkerr.pl DO NOT EDIT
- * Copyright 1995-2019 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 1995-2020 The OpenSSL Project Authors. All Rights Reserved.
  *
  * Licensed under the OpenSSL license (the "License").  You may not use
  * this file except in compliance with the License.  You can obtain a copy
@@ -341,6 +341,7 @@ static const ERR_STRING_DATA EC_str_reasons[] = {
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_LADDER_POST_FAILURE), "ladder post failure"},
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_LADDER_PRE_FAILURE), "ladder pre failure"},
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_LADDER_STEP_FAILURE), "ladder step failure"},
+    {ERR_PACK(ERR_LIB_EC, 0, EC_R_MISSING_OID), "missing OID"},
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_MISSING_PARAMETERS), "missing parameters"},
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_MISSING_PRIVATE_KEY), "missing private key"},
     {ERR_PACK(ERR_LIB_EC, 0, EC_R_NEED_NEW_SETUP_VALUES),
diff --git a/crypto/ec/ec_key.c b/crypto/ec/ec_key.c
index 08aaac5d8a6f..23efbd015ca4 100644
--- a/crypto/ec/ec_key.c
+++ b/crypto/ec/ec_key.c
@@ -1,5 +1,5 @@
 /*
- * Copyright 2002-2018 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 2002-2020 The OpenSSL Project Authors. All Rights Reserved.
  * Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved
  *
  * Licensed under the OpenSSL license (the "License").  You may not use
@@ -14,6 +14,7 @@
 #include "internal/refcount.h"
 #include <openssl/err.h>
 #include <openssl/engine.h>
+#include "crypto/bn.h"
 
 EC_KEY *EC_KEY_new(void)
 {
@@ -416,17 +417,86 @@ const BIGNUM *EC_KEY_get0_private_key(const EC_KEY *key)
 
 int EC_KEY_set_private_key(EC_KEY *key, const BIGNUM *priv_key)
 {
+    int fixed_top;
+    const BIGNUM *order = NULL;
+    BIGNUM *tmp_key = NULL;
+
     if (key->group == NULL || key->group->meth == NULL)
         return 0;
+
+    /*
+     * Not only should key->group be set, but it should also be in a valid
+     * fully initialized state.
+     *
+     * Specifically, to operate in constant time, we need that the group order
+     * is set, as we use its length as the fixed public size of any scalar used
+     * as an EC private key.
+     */
+    order = EC_GROUP_get0_order(key->group);
+    if (order == NULL || BN_is_zero(order))
+        return 0; /* This should never happen */
+
     if (key->group->meth->set_private != NULL
         && key->group->meth->set_private(key, priv_key) == 0)
         return 0;
     if (key->meth->set_private != NULL
         && key->meth->set_private(key, priv_key) == 0)
         return 0;
+
+    /*
+     * We should never leak the bit length of the secret scalar in the key,
+     * so we always set the `BN_FLG_CONSTTIME` flag on the internal `BIGNUM`
+     * holding the secret scalar.
+     *
+     * This is important also because `BN_dup()` (and `BN_copy()`) do not
+     * propagate the `BN_FLG_CONSTTIME` flag from the source `BIGNUM`, and
+     * this brings an extra risk of inadvertently losing the flag, even when
+     * the caller specifically set it.
+     *
+     * The propagation has been turned on and off a few times in the past
+     * years because in some conditions has shown unintended consequences in
+     * some code paths, so at the moment we can't fix this in the BN layer.
+     *
+     * In `EC_KEY_set_private_key()` we can work around the propagation by
+     * manually setting the flag after `BN_dup()` as we know for sure that
+     * inside the EC module the `BN_FLG_CONSTTIME` is always treated
+     * correctly and should not generate unintended consequences.
+     *
+     * Setting the BN_FLG_CONSTTIME flag alone is never enough, we also have
+     * to preallocate the BIGNUM internal buffer to a fixed public size big
+     * enough that operations performed during the processing never trigger
+     * a realloc which would leak the size of the scalar through memory
+     * accesses.
+     *
+     * Fixed Length
+     * ------------
+     *
+     * The order of the large prime subgroup of the curve is our choice for
+     * a fixed public size, as that is generally the upper bound for
+     * generating a private key in EC cryptosystems and should fit all valid
+     * secret scalars.
+     *
+     * For preallocating the BIGNUM storage we look at the number of "words"
+     * required for the internal representation of the order, and we
+     * preallocate 2 extra "words" in case any of the subsequent processing
+     * might temporarily overflow the order length.
+     */
+    tmp_key = BN_dup(priv_key);
+    if (tmp_key == NULL)
+        return 0;
+
+    BN_set_flags(tmp_key, BN_FLG_CONSTTIME);
+
+    fixed_top = bn_get_top(order) + 2;
+    if (bn_wexpand(tmp_key, fixed_top) == NULL) {
+        BN_clear_free(tmp_key);
+        return 0;
+    }
+
     BN_clear_free(key->priv_key);
-    key->priv_key = BN_dup(priv_key);
-    return (key->priv_key == NULL) ? 0 : 1;
+    key->priv_key = tmp_key;
+
+    return 1;
 }
 
 const EC_POINT *EC_KEY_get0_public_key(const EC_KEY *key)
@@ -494,6 +564,13 @@ void EC_KEY_clear_flags(EC_KEY *key, int flags)
     key->flags &= ~flags;
 }
 
+int EC_KEY_decoded_from_explicit_params(const EC_KEY *key)
+{
+    if (key == NULL || key->group == NULL)
+        return -1;
+    return key->group->decoded_from_explicit_params;
+}
+
 size_t EC_KEY_key2buf(const EC_KEY *key, point_conversion_form_t form,
                         unsigned char **pbuf, BN_CTX *ctx)
 {
diff --git a/crypto/ec/ec_lib.c b/crypto/ec/ec_lib.c
index 6832383cad51..08db89fceeb5 100644
--- a/crypto/ec/ec_lib.c
+++ b/crypto/ec/ec_lib.c
@@ -211,6 +211,7 @@ int EC_GROUP_copy(EC_GROUP *dest, const EC_GROUP *src)
 
     dest->asn1_flag = src->asn1_flag;
     dest->asn1_form = src->asn1_form;
+    dest->decoded_from_explicit_params = src->decoded_from_explicit_params;
 
     if (src->seed) {
         OPENSSL_free(dest->seed);
diff --git a/crypto/ec/ec_local.h b/crypto/ec/ec_local.h
index e656fbd5e775..64725a9c92f4 100644
--- a/crypto/ec/ec_local.h
+++ b/crypto/ec/ec_local.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2001-2019 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 2001-2020 The OpenSSL Project Authors. All Rights Reserved.
  * Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved
  *
  * Licensed under the OpenSSL license (the "License").  You may not use
@@ -209,6 +209,8 @@ struct ec_group_st {
     BIGNUM *order, *cofactor;
     int curve_name;             /* optional NID for named curve */
     int asn1_flag;              /* flag to control the asn1 encoding */
+    int decoded_from_explicit_params; /* set if decoded from explicit
+                                       * curve parameters encoding */
     point_conversion_form_t asn1_form;
     unsigned char *seed;        /* optional seed for parameters (appears in
                                  * ASN1) */
diff --git a/crypto/ec/ecp_nistp224.c b/crypto/ec/ecp_nistp224.c
index 9a9ced8f1343..6f7d66c8bea4 100644
--- a/crypto/ec/ecp_nistp224.c
+++ b/crypto/ec/ecp_nistp224.c
@@ -72,6 +72,7 @@ typedef uint64_t u64;
  */
 
 typedef uint64_t limb;
+typedef uint64_t limb_aX __attribute((__aligned__(1)));
 typedef uint128_t widelimb;
 
 typedef limb felem[4];
@@ -307,10 +308,10 @@ const EC_METHOD *EC_GFp_nistp224_method(void)
  */
 static void bin28_to_felem(felem out, const u8 in[28])
 {
-    out[0] = *((const uint64_t *)(in)) & 0x00ffffffffffffff;
-    out[1] = (*((const uint64_t *)(in + 7))) & 0x00ffffffffffffff;
-    out[2] = (*((const uint64_t *)(in + 14))) & 0x00ffffffffffffff;
-    out[3] = (*((const uint64_t *)(in+20))) >> 8;
+    out[0] = *((const limb *)(in)) & 0x00ffffffffffffff;
+    out[1] = (*((const limb_aX *)(in + 7))) & 0x00ffffffffffffff;
+    out[2] = (*((const limb_aX *)(in + 14))) & 0x00ffffffffffffff;
+    out[3] = (*((const limb_aX *)(in + 20))) >> 8;
 }
 
 static void felem_to_bin28(u8 out[28], const felem in)
diff --git a/crypto/ec/ecp_nistp521.c b/crypto/ec/ecp_nistp521.c
index 75eeba853679..08b32787293b 100644
--- a/crypto/ec/ecp_nistp521.c
+++ b/crypto/ec/ecp_nistp521.c
@@ -128,6 +128,7 @@ static const felem_bytearray nistp521_curve_params[5] = {
 # define NLIMBS 9
 
 typedef uint64_t limb;
+typedef limb limb_aX __attribute((__aligned__(1)));
 typedef limb felem[NLIMBS];
 typedef uint128_t largefelem[NLIMBS];
 
@@ -141,14 +142,14 @@ static const limb bottom58bits = 0x3ffffffffffffff;
 static void bin66_to_felem(felem out, const u8 in[66])
 {
     out[0] = (*((limb *) & in[0])) & bottom58bits;
-    out[1] = (*((limb *) & in[7]) >> 2) & bottom58bits;
-    out[2] = (*((limb *) & in[14]) >> 4) & bottom58bits;
-    out[3] = (*((limb *) & in[21]) >> 6) & bottom58bits;
-    out[4] = (*((limb *) & in[29])) & bottom58bits;
-    out[5] = (*((limb *) & in[36]) >> 2) & bottom58bits;
-    out[6] = (*((limb *) & in[43]) >> 4) & bottom58bits;
-    out[7] = (*((limb *) & in[50]) >> 6) & bottom58bits;
-    out[8] = (*((limb *) & in[58])) & bottom57bits;
+    out[1] = (*((limb_aX *) & in[7]) >> 2) & bottom58bits;
+    out[2] = (*((limb_aX *) & in[14]) >> 4) & bottom58bits;
+    out[3] = (*((limb_aX *) & in[21]) >> 6) & bottom58bits;
+    out[4] = (*((limb_aX *) & in[29])) & bottom58bits;
+    out[5] = (*((limb_aX *) & in[36]) >> 2) & bottom58bits;
+    out[6] = (*((limb_aX *) & in[43]) >> 4) & bottom58bits;
+    out[7] = (*((limb_aX *) & in[50]) >> 6) & bottom58bits;
+    out[8] = (*((limb_aX *) & in[58])) & bottom57bits;
 }
 
 /*
@@ -159,14 +160,14 @@ static void felem_to_bin66(u8 out[66], const felem in)
 {
     memset(out, 0, 66);
     (*((limb *) & out[0])) = in[0];
-    (*((limb *) & out[7])) |= in[1] << 2;
-    (*((limb *) & out[14])) |= in[2] << 4;
-    (*((limb *) & out[21])) |= in[3] << 6;
-    (*((limb *) & out[29])) = in[4];
-    (*((limb *) & out[36])) |= in[5] << 2;
-    (*((limb *) & out[43])) |= in[6] << 4;
-    (*((limb *) & out[50])) |= in[7] << 6;
-    (*((limb *) & out[58])) = in[8];
+    (*((limb_aX *) & out[7])) |= in[1] << 2;
+    (*((limb_aX *) & out[14])) |= in[2] << 4;
+    (*((limb_aX *) & out[21])) |= in[3] << 6;
+    (*((limb_aX *) & out[29])) = in[4];
+    (*((limb_aX *) & out[36])) |= in[5] << 2;
+    (*((limb_aX *) & out[43])) |= in[6] << 4;
+    (*((limb_aX *) & out[50])) |= in[7] << 6;
+    (*((limb_aX *) & out[58])) = in[8];
 }
 
 /* BN_to_felem converts an OpenSSL BIGNUM into an felem */
diff --git a/crypto/ec/ecp_nistz256.c b/crypto/ec/ecp_nistz256.c
index ba9268138862..5005249b05ea 100644
--- a/crypto/ec/ecp_nistz256.c
+++ b/crypto/ec/ecp_nistz256.c
@@ -929,207 +929,6 @@ __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
     return ret;
 }
 
-/*
- * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
- * code processing 4 points in parallel, corresponding serial operation
- * is several times slower, because it uses 29x29=58-bit multiplication
- * as opposite to 64x64=128-bit in integer-only scalar case. As result
- * it doesn't provide *significant* performance improvement. Note that
- * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
- * you'd need to compile even asm/ecp_nistz256-avx.pl module.
- */
-#if defined(ECP_NISTZ256_AVX2)
-# if !(defined(__x86_64) || defined(__x86_64__) || \
-       defined(_M_AMD64) || defined(_M_X64)) || \
-     !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
-#  undef ECP_NISTZ256_AVX2
-# else
-/* Constant time access, loading four values, from four consecutive tables */
-void ecp_nistz256_avx2_multi_gather_w7(void *result, const void *in,
-                                       int index0, int index1, int index2,
-                                       int index3);
-void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in);
-void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4);
-void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4,
-                                           const void *Bx4);
-void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4,
-                                            const void *Bx4);
-void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4);
-void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4);
-void ecp_nistz256_avx2_set1(void *RESULTx4);
-int ecp_nistz_avx2_eligible(void);
-
-static void booth_recode_w7(unsigned char *sign,
-                            unsigned char *digit, unsigned char in)
-{
-    unsigned char s, d;
-
-    s = ~((in >> 7) - 1);
-    d = (1 << 8) - in - 1;
-    d = (d & s) | (in & ~s);
-    d = (d >> 1) + (d & 1);
-
-    *sign = s & 1;
-    *digit = d;
-}
-
-/*
- * ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
- * precomputed table. It does 4 affine point additions in parallel,
- * significantly speeding up point multiplication for a fixed value.
- */
-static void ecp_nistz256_avx2_mul_g(P256_POINT *r,
-                                    unsigned char p_str[33],
-                                    const P256_POINT_AFFINE(*preComputedTable)[64])
-{
-    const unsigned int window_size = 7;
-    const unsigned int mask = (1 << (window_size + 1)) - 1;
-    unsigned int wvalue;
-    /* Using 4 windows at a time */
-    unsigned char sign0, digit0;
-    unsigned char sign1, digit1;
-    unsigned char sign2, digit2;
-    unsigned char sign3, digit3;
-    unsigned int idx = 0;
-    BN_ULONG tmp[P256_LIMBS];
-    int i;
-
-    ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 };
-    ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 };
-    ALIGN32 P256_POINT_AFFINE point_arr[4];
-    ALIGN32 P256_POINT res_point_arr[4];
-
-    /* Initial four windows */
-    wvalue = *((u16 *) & p_str[0]);
-    wvalue = (wvalue << 1) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign0, &digit0, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign1, &digit1, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign2, &digit2, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign3, &digit3, wvalue);
-
-    ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[0],
-                                      digit0, digit1, digit2, digit3);
-
-    ecp_nistz256_neg(tmp, point_arr[0].Y);
-    copy_conditional(point_arr[0].Y, tmp, sign0);
-    ecp_nistz256_neg(tmp, point_arr[1].Y);
-    copy_conditional(point_arr[1].Y, tmp, sign1);
-    ecp_nistz256_neg(tmp, point_arr[2].Y);
-    copy_conditional(point_arr[2].Y, tmp, sign2);
-    ecp_nistz256_neg(tmp, point_arr[3].Y);
-    copy_conditional(point_arr[3].Y, tmp, sign3);
-
-    ecp_nistz256_avx2_transpose_convert(aX4, point_arr);
-    ecp_nistz256_avx2_to_mont(aX4, aX4);
-    ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]);
-    ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]);
-
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign0, &digit0, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign1, &digit1, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign2, &digit2, wvalue);
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    idx += window_size;
-    booth_recode_w7(&sign3, &digit3, wvalue);
-
-    ecp_nistz256_avx2_multi_gather_w7(point_arr, preComputedTable[4 * 1],
-                                      digit0, digit1, digit2, digit3);
-
-    ecp_nistz256_neg(tmp, point_arr[0].Y);
-    copy_conditional(point_arr[0].Y, tmp, sign0);
-    ecp_nistz256_neg(tmp, point_arr[1].Y);
-    copy_conditional(point_arr[1].Y, tmp, sign1);
-    ecp_nistz256_neg(tmp, point_arr[2].Y);
-    copy_conditional(point_arr[2].Y, tmp, sign2);
-    ecp_nistz256_neg(tmp, point_arr[3].Y);
-    copy_conditional(point_arr[3].Y, tmp, sign3);
-
-    ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
-    ecp_nistz256_avx2_to_mont(bX4, bX4);
-    ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
-    /* Optimized when both inputs are affine */
-    ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4);
-
-    for (i = 2; i < 9; i++) {
-        wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-        wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-        idx += window_size;
-        booth_recode_w7(&sign0, &digit0, wvalue);
-        wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-        wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-        idx += window_size;
-        booth_recode_w7(&sign1, &digit1, wvalue);
-        wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-        wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-        idx += window_size;
-        booth_recode_w7(&sign2, &digit2, wvalue);
-        wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-        wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-        idx += window_size;
-        booth_recode_w7(&sign3, &digit3, wvalue);
-
-        ecp_nistz256_avx2_multi_gather_w7(point_arr,
-                                          preComputedTable[4 * i],
-                                          digit0, digit1, digit2, digit3);
-
-        ecp_nistz256_neg(tmp, point_arr[0].Y);
-        copy_conditional(point_arr[0].Y, tmp, sign0);
-        ecp_nistz256_neg(tmp, point_arr[1].Y);
-        copy_conditional(point_arr[1].Y, tmp, sign1);
-        ecp_nistz256_neg(tmp, point_arr[2].Y);
-        copy_conditional(point_arr[2].Y, tmp, sign2);
-        ecp_nistz256_neg(tmp, point_arr[3].Y);
-        copy_conditional(point_arr[3].Y, tmp, sign3);
-
-        ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
-        ecp_nistz256_avx2_to_mont(bX4, bX4);
-        ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
-
-        ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4);
-    }
-
-    ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]);
-    ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]);
-    ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]);
-
-    ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4);
-    /* Last window is performed serially */
-    wvalue = *((u16 *) & p_str[(idx - 1) / 8]);
-    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-    booth_recode_w7(&sign0, &digit0, wvalue);
-    ecp_nistz256_gather_w7((P256_POINT_AFFINE *)r,
-                           preComputedTable[36], digit0);
-    ecp_nistz256_neg(tmp, r->Y);
-    copy_conditional(r->Y, tmp, sign0);
-    memcpy(r->Z, ONE, sizeof(ONE));
-    /* Sum the four windows */
-    ecp_nistz256_point_add(r, r, &res_point_arr[0]);
-    ecp_nistz256_point_add(r, r, &res_point_arr[1]);
-    ecp_nistz256_point_add(r, r, &res_point_arr[2]);
-    ecp_nistz256_point_add(r, r, &res_point_arr[3]);
-}
-# endif
-#endif
-
 __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
                                                const P256_POINT_AFFINE *in,
                                                BN_CTX *ctx)
@@ -1219,6 +1018,8 @@ __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
         }
 
         if (preComputedTable) {
+            BN_ULONG infty;
+
             if ((BN_num_bits(scalar) > 256)
                 || BN_is_negative(scalar)) {
                 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
@@ -1250,67 +1051,58 @@ __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
             for (; i < 33; i++)
                 p_str[i] = 0;
 
-#if defined(ECP_NISTZ256_AVX2)
-            if (ecp_nistz_avx2_eligible()) {
-                ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable);
-            } else
-#endif
-            {
-                BN_ULONG infty;
+            /* First window */
+            wvalue = (p_str[0] << 1) & mask;
+            idx += window_size;
 
-                /* First window */
-                wvalue = (p_str[0] << 1) & mask;
-                idx += window_size;
+            wvalue = _booth_recode_w7(wvalue);
 
-                wvalue = _booth_recode_w7(wvalue);
+            ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
+                                   wvalue >> 1);
 
-                ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
-                                       wvalue >> 1);
-
-                ecp_nistz256_neg(p.p.Z, p.p.Y);
-                copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
-
-                /*
-                 * Since affine infinity is encoded as (0,0) and
-                 * Jacobian ias (,,0), we need to harmonize them
-                 * by assigning "one" or zero to Z.
-                 */
-                infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] |
-                         p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
-                if (P256_LIMBS == 8)
-                    infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] |
-                              p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
-
-                infty = 0 - is_zero(infty);
-                infty = ~infty;
-
-                p.p.Z[0] = ONE[0] & infty;
-                p.p.Z[1] = ONE[1] & infty;
-                p.p.Z[2] = ONE[2] & infty;
-                p.p.Z[3] = ONE[3] & infty;
-                if (P256_LIMBS == 8) {
-                    p.p.Z[4] = ONE[4] & infty;
-                    p.p.Z[5] = ONE[5] & infty;
-                    p.p.Z[6] = ONE[6] & infty;
-                    p.p.Z[7] = ONE[7] & infty;
-                }
+            ecp_nistz256_neg(p.p.Z, p.p.Y);
+            copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
 
-                for (i = 1; i < 37; i++) {
-                    unsigned int off = (idx - 1) / 8;
-                    wvalue = p_str[off] | p_str[off + 1] << 8;
-                    wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
-                    idx += window_size;
+            /*
+             * Since affine infinity is encoded as (0,0) and
+             * Jacobian is (,,0), we need to harmonize them
+             * by assigning "one" or zero to Z.
+             */
+            infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] |
+                     p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
+            if (P256_LIMBS == 8)
+                infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] |
+                          p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
+
+            infty = 0 - is_zero(infty);
+            infty = ~infty;
+
+            p.p.Z[0] = ONE[0] & infty;
+            p.p.Z[1] = ONE[1] & infty;
+            p.p.Z[2] = ONE[2] & infty;
+            p.p.Z[3] = ONE[3] & infty;
+            if (P256_LIMBS == 8) {
+                p.p.Z[4] = ONE[4] & infty;
+                p.p.Z[5] = ONE[5] & infty;
+                p.p.Z[6] = ONE[6] & infty;
+                p.p.Z[7] = ONE[7] & infty;
+            }
 
-                    wvalue = _booth_recode_w7(wvalue);
+            for (i = 1; i < 37; i++) {
+                unsigned int off = (idx - 1) / 8;
+                wvalue = p_str[off] | p_str[off + 1] << 8;
+                wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
+                idx += window_size;
 
-                    ecp_nistz256_gather_w7(&t.a,
-                                           preComputedTable[i], wvalue >> 1);
+                wvalue = _booth_recode_w7(wvalue);
 
-                    ecp_nistz256_neg(t.p.Z, t.a.Y);
-                    copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
+                ecp_nistz256_gather_w7(&t.a,
+                                       preComputedTable[i], wvalue >> 1);
 
-                    ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
-                }
+                ecp_nistz256_neg(t.p.Z, t.a.Y);
+                copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
+
+                ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
             }
         } else {
             p_is_infinity = 1;